Wolfgang Klimesch (1995) Memory Processes Described as Brain Oscillations. Psycoloquy: 6(06) Memory Brain (1)

Volume: 6 (next, prev) Issue: 06 (next, prev) Article: 1 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 6(06): Memory Processes Described as Brain Oscillations


Wolfgang Klimesch
University of Salzburg
Department of Physiological Psychology
Institute of Psychology, Hellbrunnerstr. 34
A-5020 Salzburg, AUSTRIA



This target article tries to integrate results in memory research from diverse disciplines such as psychophysiology, cognitive psychology, anatomy and neurophysiology. The integrating link is seen in more recent anatomical findings that provide strong arguments for the assumption that oscillations provide the basic form of communication between cortical cell assemblies. The basic argument is that episodic memory processes, which are part of a complex working memory system, are reflected by oscillations in the theta band, whereas long-term memory processes are reflected by alpha oscillations. It is assumed that alpha and theta oscillations serve to encode, access, and retrieve cortical codes that are stored in the form of widely distributed but intensely interconnected cell assemblies.


Alpha, EEG, Hippocampus, Memory, Oscillation, Thalamus, Theta.


1. In the following sections an attempt is made to link convergent knowledge about memory processes from four different fields: electrophysiology, cognitive psychology, functional anatomy, and neurophysiology. The motivation to follow up this interdisciplinary approach was spurred by the following two basic findings.

2. The first finding is that the frequency of EEG-alpha oscillations is positively correlated with memory performance. In a series of four independent experiments, we (Klimesch, Schimke, Ladurner & Pfurtscheller, 1990; Klimesch, Schimke & Pfurtscheller, 1993; and Klimesch, Schimke, Doppelmayr, Ripper, Schwaiger & Pfurtscheller, 1994) were able to demonstrate that good memory performers have a significantly higher alpha frequency than subjects with bad memory performance. In concordance with these results, reaction time experiments have also shown that with increasing memory performance, retrieval time decreases (Klimesch, Schimke & Ladurner, 1988).

3. The second finding stems from reaction time experiments with natural concepts. In a variety of different experiments (summarized in Klimesch, 1994), it was found that complex concepts can be processed much faster than less complex concepts and it could be demonstrated that this effect is not due to confounded variables such as typicality, word frequency or degree of connectivity. For cognitive psychology, this fact that more complex information can be processed faster than less complex information is a challenge, because well established theories (such as the well known ACT* theory from Anderson, 1983) and related experimental results show the opposite effect. In dealing with this challenge a new model, the connectivity model of semantic processing, was developed (Klimesch, 1987, 1994). The connectivity model focuses on an explicit description of representational assumptions for the encoding of long-term memory (LTM) codes and describes semantic search and retrieval processes in terms of a spreading activation process. It is important to emphasize that this positive speed effect is predicted and holds true for integrated (interconnected) semantic LTM codes but not for episodic short-term memory (STM) demands (Kroll & Klimesch, 1992).

4. Both findings point towards the importance of what will be termed the speed effect of memory performance: High memory performance and complex integrated knowledge speed up search and retrieval processes. With respect to the electrophysiological results, the interesting fact is that the connectivity model explains LTM performance in terms of an increase in the speed of the spreading activation process. It thus seems plausible to see a possible link with the finding that alpha frequency increases with increasing memory performance. Because there is evidence that EEG-alpha activity is related to thalamo-cortical oscillations (e.g., Steriade, Jones & Llinas, 1990) it is tempting to postulate a specific role of the thalamo-cortical network for memory processes.

5. In trying to integrate these results within a psychophysiological perspective, we suggest the following preliminary hypothesis which rests on three assumptions: (1) Memory codes are stored in the form of interconnected but widely distributed networks (cell assemblies) in the cortex. (2) Memory codes are accessed and retrieved via "longitudinal" pathways linking deeper brain structures such as the thalamus with the cortex. And (3) alpha is one of the dominant rhythms reflecting the activity of some of these pathways. This hypothesis has one crucial implication and raises several questions. The most important questions are:

  - What is the neuro-anatomical and physiological basis for
    memory processes in general? (see section IV below).

  - Are different types of memory processes reflected by different
    types of oscillations? (see sections V and VI below).

6. The crucial implication of the proposed hypothesis is that brain oscillations may be considered the basic phenomenon of cortical information processing. It should be noted that this implication is already inherent to the finding of a positive relationship between EEG-alpha frequency and memory performance. Because dominant brain oscillations can be recorded with the EEG, a summary of basic results is described in sections II.2 and II.3 below. These sections will provide us with the necessary conceptual tools to describe memory processes in terms of oscillations.

7. EEG findings that focus on the analysis of brain rhythms will be an important source of evidence for evaluating the proposed hypothesis. For reasons that are explained later in the text it will be assumed that EEG-alpha oscillations are closely related to semantic long-term memory (LTM) processes. Thus, when contrasting LTM with STM processes, the interesting question arises: Which frequency band in the EEG might be related to the encoding and retrieval of new information? It is well known that on the one hand, the hippocampal formation is of crucial importance for the encoding of new information and that on the other hand, the theta rhythm is the dominant oscillation of this brain region. It is, therefore, tempting to expect a specific relationship between the theta rhythm and the encoding and retrieval of new information.

8. Besides some clinical findings (Arnolds, Lopes da Silva, Aitink, Kamp & Boeijinga, 1980) and recent results from our laboratory (Klimesch, Schimke & Schwaiger, 1994), there is a lack of clear-cut findings that point towards the proposed relationship between theta frequency and episodic STM demands in humans. This lack most likely is due to the fact that theta is not a dominant rhythm in the EEG of wake adults (see section V). Most of the results in the literature that deal with EEG and STM were reported in studies using event-related brain potentials (ERPs). The fact that there is no obvious link between ERP findings and brain oscillations brings us to the most speculative part of the manuscript. In a later section it is proposed that ERP-components that reflect certain STM demands may be due to phase locked theta activity (see section VI).


9. The basic idea proposed in this and the following sections is that memory processes such as search, spreading activation, and retrieval can be described as processes that modulate the frequency of oscillatory neuronal discharge patterns. For sensory information processing it is well established that it is the modulation of the frequency of action potentials that encodes the information of a sensory input. This fact will be applied to cortical information processing. It will be assumed that the modulation of the frequency of brain oscillations is the basic mechanism of information transmission in the cortex.


10. Braitenberg and his coworkers have shown that some of the conventional ideas about the anatomy of the cortex are wrong (see the comprehensive review in Braitenberg & Schuez, 1991). The question he and his group addresses refers to the issue of specificity or randomness of neuronal connections in the cortex. They demonstrate that the probability for an axonal synapse to have a particular neuron as postsynaptic partner is p = .001. The probability that more than one contact is made with a particular cell is extremely small (e.g., for three contacts the probability is p = .0000001). Given this enormous divergence in interconnectedness, one cell could never excite any other neuron. Due to the principle of temporal and spatial summation (summarized, e.g., in Koestner, 1985), a single cell will respond with an outgoing signal (an action potential) only if many convergent inputs are received at the same time (within a narrow time window). This means that converging neuronal signals must arrive synchronously in order to trigger an outgoing signal.

11. The crucial question now is, which mechanisms operate to synchronize the neuronal input for each neuron? It is proposed that oscillations reflect this synchronizing mechanism. If signals come in synchronized bursts of action potentials, that is, in the form of oscillations, single neurons even in a distributed, randomly wired network will respond with an outgoing signal. Oscillations may be induced into a neuronal network by pacemaker cells and/or by endogenous membrane properties of individual cells (e.g., the reviews in Steriade, Jones & Llinas, 1990; and Basar & Bullock, 1992). It is interesting to note that mathematical analyses indicate that, particularly in biological systems, oscillators tend to synchronize if their frequencies are not too different from each other (Strogatz & Stewart, 1993).


12. One of the best known results obtained with the EEG documents the functional importance of brain oscillations. Since the pioneering work of Berger in the late 1920s and early 1930s (e.g., Berger, 1929), it is known that the most dominant rhythm in the EEG, the alpha rhythm, can best be seen under conditions of relative mental inactivity, but it is blocked or desynchronized by attention and/or mental effort. The fact that a strong rhythm, such as the alpha rhythm, can be recorded from scalp electrodes means that millions of cortical neurons must oscillate synchronously with the same phase and within a comparatively narrow frequency band. Desynchronization, that is, the disappearance of the dominant alpha rhythm is functionally related to mental activity and means that different oscillators within the alpha band are no longer coupled. They oscillate with different phase lags and probably with different frequencies. This basic EEG-phenomenon of synchronization (during mental inactivity) and desynchronization (during mental activity) provides us with a preliminary but nonetheless important understanding of how information may be processed in the brain: The synchronization of very large populations of neurons oscillating with the same phase and frequency reflects a state in which no information is transmitted.


13. It is of crucial importance to emphasize that synchronization has two different meanings. Synchronization within the traditional context of EEG research reflects a state of mental inactivity. More recent research with microelectrodes implanted in the cortex, however, have shown that synchronous oscillatory discharge patterns in high frequency bands (such as the broad gamma band from 30 - 70 Hz) are related to rather localized cortical processes, reflecting cognitive activity such as visual encoding processes (e.g., Gray & Singer, 1987). Only at first glance do these two different meanings of neuronal synchronization seem to contradict each other. From the standpoint of EEG research, it is a matter of resolution whether or not we may speak of synchronization or desynchronization. Desynchronization of the EEG is interpreted in terms of frequency and/or phase shifts of a large population of oscillators that become progressively uncoupled. Thus, recorded from EEG macroelectrodes, neuronal activity appears desynchronized. Nonetheless, within small cortical areas, neuronal activity may still exhibit a synchronous discharge pattern. To avoid confusion, we will call the synchronous activity of large cortical areas reflecting mental inactivity type 1 synchronization. With type 2 synchronization we denote the synchronous oscillatory discharge pattern of selected and comparatively small cortical areas.

14. In summarizing these three terms, we have to keep in mind that on the microscale, a frequency change (possibly over a broad frequency range) of a comparatively small group of neurons (or cell assembly) occurring synchronously within the different neurons of this group (cell assembly) reflects the processing of information (see also section III.2). Because information processing generally is considered a distributed process, a great number of different, distributed cell assemblies will show type 2 synchronization in response to cognitive demands. On the macroscale, however, the behavior of many different cell assemblies responding with type 2 synchronization will show up as desynchronization in the EEG. The main reason for this is that each cell assembly will respond with its own frequency and this synchronization may not be coupled between cell assemblies. Because the synchronous discharge of small cell assemblies is a very weak signal for the EEG, type 2 synchronization can be detected primarily by microelectrodes but is difficult to detect with EEG macroelectrodes. Thus, if many different cell assemblies show uncoupled type 2 synchronization, the EEG will be de-synchronized. In contrast, type 1 synchronization is a very strong signal for the EEG, showing the synchronous, phase coupled oscillatory discharge pattern within a narrow frequency band of very large cell populations.

15. EEG-frequencies are conventionally subdivided in frequency bands such as the theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (14 to about 30 Hz) and gamma bands (30 - 70 Hz). It is important to note that the traditional terms of EEG (type 1) synchronization and desynchronization apply for the alpha and beta bands only. The gamma band seems to synchronize in response to cognitive demands (Pfurtscheller, Flotzinger & Neuper, 1994) and seems to reflect real type 2 synchronization in the EEG (see Pulvermueller et al., 1994)., The theta band clearly synchronizes in response to cognitive demands.

16. This basic behavior of the EEG generally is very similar for animals and humans (see e.g., the review in Lopes da Silva, 1992) with the exception that the frequency range of the theta rhythm is much wider in animals (lower mammals). Thus, to avoid confusions with the human EEG, the term rhythmic slow activity (RSA) was introduced to denote synchronized theta activity (see, e.g., Vanderwolf & Robinson, 1981) whereas the term irregular slow activity (ISA) is used to denote desynchronized theta activity. In contrast to ISA, RSA reflects a state of mental activity in animals. In humans too, theta synchronization is related to mental activity and to the encoding of episodic information in particular (see Klimesch, Schimke & Schwaiger, 1994; and Arnolds et al., 1980).

17. In the human EEG of wake adults, theta is a weak rhythm that most likely is induced into the cortex via a small but distributed set of longitudinal hippocampo-cortical pathways (see sections III, V and the review in Lopes da Silva, 1992). In wake adults, theta synchronization seems to have the meaning of coupled type 2 synchronization. This sort of synchronization is explained in terms of a small subset of hippocampo-cortical feedback loops responding to an appropriate event or signal with synchronized phase locked theta activity. As a result, selected and distributed cortical cell assemblies will start to respond with synchronous theta activity. According to this interpretation, theta desynchronization (ISA in animals) simply is the lack of type 2 synchronization (RSA in animals). It is well known, however, that during certain sleep stages, theta becomes a dominant rhythm in the EEG, reflecting type 1 synchronization. Note that this is not in contradiction to the proposed interpretation: Type 2 theta synchronization reflects the processing of information whereas type 1 theta synchronization reflects the lack of information processing or a state of "functional inhibition".

18. A good example to document the meaningfulness of type 1 synchronization for memory processes is the general issue of inhibition. Any memory theory trying to explain search processes is confronted with the fundamental problem of how it can be explained that spreading activation is confined to the relevant parts of the (cortical) memory network (Klimesch, 1994). The most obvious way of handling this problem is to assume strong inhibitory processes that allow a search process to spread only within certain regions of the network. According to Braitenberg & Schuez (1991), however, the assumption of powerful inhibitory processes is not plausible, given the fact that about 85% of all cortical neurons are pyramidal cells with excitatory synapses. Inhibitory synapses are comparatively rare (comprising only 11 to 15% of all cortical synapses) and reside on stellate cells that primarily make local connections. Thus, in the cortical network, inhibitory processes are more likely operating locally and probably do not have far reaching effects. Synchronous (type 1) oscillations within a narrow frequency band, induced in large areas or even in the entire cortex (e.g., in sleep), may have powerful inhibitory effects. When synchronous (type 1) oscillations within a narrow frequency band are selectively induced in certain cortical areas from a part in the brain that operates as some kind of "monitoring unit" or "control unit", the basic theoretical framework for explaining inhibitory processes by type 1 synchronization is at hand. According to this idea, type 1 synchronization could act to block a search process (reflected by type 2 synchronization) from entering irrelevant parts of the network.



19. Two basic aspects of memory processes are emphasized. The one refers to a close interaction between the working memory system (WMS, see, e.g., Baddeley, 1992) with the long-term memory system (LTMS). This interaction plays an important role for encoding, searching, retrieving, and recognizing information. The other aspect refers to the meaning of monitoring processes which are considered a heterogeneous class of processes within the WMS that operate under voluntary control. The close interaction between the WMS and LTMS can be demonstrated by considering a fundamental cognitive process such as recognizing a familiar object. The crucial idea here is that after a sensory code is established, bottom up processes access semantic information in LTM that is used to identify the perceived object. If the process of identification which is considered a matching processes yields a positive result, the object is recognized which in turn leads to the creation of a STM code. The matching process activates pathways that are similar or identical to those that serve to retrieve information from LTM. Top down processes which are guided by expectancy and selective attention are capable of directing the matching process towards a certain outcome by preactivating or preselecting appropriate templates or prototypes in LTM. Recent models, such as Grossberg's adaptive resonance theory (ART), also proceed from the basic assumption that templates or prototypes stored in LTM are activated during a matching process which is characterized by a close interaction between STM and LTM (e.g., Grossberg, 1980; Grossberg & Stone, 1986; Carpenter & Grossberg, 1993).

20. Note that encoding has two different meanings. The encoding of sensory information (as a process of recognition) aims at the semantic understanding of perceived information. LTM holds that information which is essential for this encoding process. Within the framework of STM, encoding means the creation of a new code that primarily comprises episodic information. According to Tulving (e.g., Tulving, 1984), episodic information is that type of contextual information which keeps an individual autobiographically oriented within space and time.


21. The necessity for monitoring processes is related to the permanent need to update episodic information. An episodic code which is created through the action of monitoring processes reflects primarily subjective information, such as context, expectancy, emotion, and certain autobiographic aspects. Because time changes the context permanently, there is a permanent need to update and store episodic information. STM serves this vital need to store episodic information within certain capacity limits. Beyond these limits episodic information may be stored into a more permanent memory system. In contrast to episodic information, the encoding of new semantic information requires in most cases special mnemo-techniques ("learning"). Because "learning" is guided by complex monitoring processes of the WMS, it is assumed that new semantic and episodic information use identical encoding pathways into LTM.

22. As the classical case of patient H.M. (Scoville & Milner, 1957; and the reviews in, e.g., Markowitsch, 1983, 1984, and Markowitsch & Pritzel, 1985, for similar cases) as well as a variety of more recent evidence demonstrates, the hippocampus (and other parts of the limbic system) are responsible for encoding (or retrieving) any new declarative information beyond the capacity limits of STM (Squire, 1992; Squire, Knowlton & Musen, 1993). Because of the permanent need to update episodic (but not semantic) information, the loss of freshly encoded episodic memory is such a dramatic symptom for anterograde amnesia that the concurrent failure to encode new semantic information appeared to be of minor significance (c.f. Baddeley's commentary to Tulving's target article in BBS for a similar statement; Baddeley, 1984, p. 239). The importance of the hippocampus for contextual encoding was proposed by several researchers. As an example, Teyler and DiScenna (1986) assume that the hippocampus stores at least initially some sort of "index" pointing towards those neocortical modules or cell assemblies (i.e., those LTM structures) that have been activated. In an interesting theory, Miller (1991) proposes that the hippocampus is important for contextual representations and might be involved in forming global cell assemblies. Squire (1992) emphasizes the "binding" function of the hippocampus and emphasizes that it is needed to bind together distributed cell assemblies (representing features) that together form the information of a single code. If the hippocampus is involved in the process of consolidation, contextual encoding and binding, we have to expect that extensive and widespread projections exist to the association cortices. This is indeed the case (e.g., Lopes da Silva, Witter, Boeijinga & Lohman, 1990).

23. These conceptions of hippocampal functions fit very well with the hypothesis (proposed in section V, paragraph 42, see also paragraph 17) that the selective type 2 synchronization of a small percentage of hippocampo-cortical feedback loops actually reflects the encoding of new (episodic) information. The synchronization of selected, distributed cortical cell assemblies might allow for establishing a binding process that links features in a new way to create a new context. Carpenter and Grossberg (1993) add another interesting aspect to this view of hippocampal functions. They assume that the hippocampus represents something like an orienting system that allows one to orient towards the encoding of a new stimulus. It is important to note that all of the functions ascribed to the hippocampus are central functions of the WMS.


24. It is assumed that LTM codes are represented by a distributed structure of nodes that establish a complex network or cell assembly in the neocortex. Even a single node is considered a structure of features that may be widely distributed throughout different regions of the neocortex. Features may be represented by smaller cell assemblies such as cortical columns or modules which serve as feature detectors when activated by perceptual processes (c.f. the close interaction between LTM and perception emphasized in paragraph 19). The assumption of highly distributed codes may explain why any attempt to localize a particular engram resulted in a failure (c.f. Lashley, 1950). This structural encoding assumption leads to the crucial question of how the different features belonging to a single code can be activated together as a functional unity without activating features of other overlapping but irrelevant codes. According to the traditional view as first proposed by Hebb (e.g., Hebb, 1949), one may assume that the features of a code are represented by a cell assembly of interconnected cells that are functionally characterized by a concurrent elevation of their average firing rate. Unlike more recent models, Hebb's conception has the disadvantage that in a particular cortical region and within a given time span, only a single code or feature can be activated, because the enhanced firing rate is the only cue which allows it to distinguish the relevant code from irrelevant information. During a search process in LTM, a huge variety of codes will be activated at the same time and possibly in the same brain region. Thus, different and topographically overlapping cell assemblies will be activated at the same time. Consequently, it will be impossible to distinguish between different codes. In trying to avoid this problem, one may instead assume that assemblies may be functionally defined by a state of synchronous firing of cortical neurons, rather than by an enhanced average firing rate. This means that in a particular cortical region and within a given time span, all of the cells may be highly active, but only those cells firing synchronously represent the relevant information comprised by a single code.

25. Gray and Singer (1987) together with other researchers at the Frankfurt MPI (e.g., Gray, Koenig, Engel & Singer, 1989; Engel, Koenig, Kreiter, Schillen & Singer, 1992) have provided convincing evidence that a visual code, established through a perceptual process, can be described as a cell assembly which responds with a synchronous oscillatory discharge pattern within a broad frequency range of about 30 to 70 Hz which is termed gamma band. They assume that the synchronous oscillatory firing pattern of distributed cortical cells reflects a stage of cortical integration in the sense that the information provided by different feature detectors is integrated into a single visual code. This assumption is substantiated by the important finding that even widely distributed but synchronously oscillating cell assemblies fire with zero phase lag. Feedback loops, connecting different cell groups of the cortex, obviously are the means which enable this surprising ability. Oscillations are considered carrier signals for the relevant information which might be encoded by the synchronous modulation of frequencies.

26. When applying the encoding principle described above to search processes in LTM, the interesting conclusion is that the search process would have to find cell assemblies that are capable of establishing a synchronous oscillatory firing pattern in response to the initiation of a search process. According to this concept, the relevant sought-after information would be characterized by a synchronous oscillatory discharge pattern. However, unlike a visual encoding process, where all the relevant information is given at the same time, during the course of a search process thousands of different codes will be activated at different times. Each code may respond with a synchronous oscillatory discharge pattern, but what should be the criterion to distinguish the sought-after relevant from irrelevant information? Establishing a synchronous oscillatory firing pattern along the search pathway would not allow the search process to selectively retrieve the relevant information.

27. This question of how a search process finds the relevant information is called the search problem. It will be explained in terms of a spreading activation process that was described within the framework of the connectivity model, outlined in detail elsewhere (Klimesch, 1994). This model describes spreading activation in abstract terms of different activation values moving from one node to another. In the context of cortical activation the term "activation value" is translated into "frequency of an oscillatory neuronal discharge pattern". As an example, let us consider a completely interconnected code (with n = 5 nodes), in which each node is connected to each of the other nodes. At the beginning of an activation process each node representing a cell assembly (or cortical module) has zero activity which means that it oscillates with some low (resting) frequency. Activating a node means to put it in a state of oscillation with a frequency that is higher than its resting frequency. Now, if activation starts at one of the n = 5 nodes with frequency f, this activation spreads to all of the other n - 1 nodes of that code. Accordingly nodes 2, 3, 4 and 5 are also put in oscillation with frequency f. Now in a second activation stage, the n - 1 nodes activate each other. Thus, each node receives activation from the remaining n - 2 nodes. With each additional activation, the n - 1 nodes increase their responsiveness which means that they increase their frequency proportional to the number of times they were activated. Note that all of the n - 1 nodes are completely interconnected and are thus n - 2 times activated which results in an increase in frequency from f to f'. In a third step, the increased frequency f' is fed back to that node where the activation process was initiated. Note that the increase from f to f' reflects the complexity of a code. The more nodes there are, the higher frequency f' and consequently, the faster the spreading activation process will be. Furthermore and most important, due to the interconnections between the n - 1 nodes, which are considered the features of a code, a synchronous oscillatory discharge pattern is established within all of the components of a code. Thus, in accordance with the findings of Gray & Singer (1987) which are partly summarized in Engel et al. (1992), a memory code can be characterized by a pattern of features oscillating synchronously. However, during the spreading activation process each code is activated at different times and, even more important, each code will respond with a different frequency, because frequency f' depends on geometric properties which differ between codes.

28. According to the connectivity model, a search process terminates with a positive result if activation (i.e., some frequency f' which must be higher then input frequency f) spreads back as "echo" to one of those nodes where the search was initiated. Monitoring processes of the WMS do not guide the spreading activation process which follows automatically by local mechanisms. Their task is to select access points, when initiating a search process, and to retrieve the relevant information if a search process terminates with a positive result. The result of a search process can be judged by the strength of activation equaling frequency f' of activated codes. That code responding with the highest frequency represents the relevant information to be retrieved.

29. It is important to remember the two different types of synchronization, outlined in section II.3. Synchronous (type 1) oscillations of large cell populations are considered to reflect a state of inhibition. As an example, at the beginning of a search process, large cortical areas may oscillate synchronously with resting frequency f. Because information is encoded by the modulation of frequencies, large cell populations oscillating with the same (low resting) frequency (within a narrow band) are not capable of transmitting information. However, as a result of a search (or spreading activation) process, different cortical cell assemblies change their frequency and establish a type 2 synchronous oscillatory discharge pattern over a broad frequency range which may comprise the entire frequency band (comprising the theta, alpha, beta and gamma bands). This modulation or change in frequency reflects the process of transmitting information but is restricted to those cortical areas that are relevant for processing a task.

30. With respect to the most dominant rhythm in the EEG, a spreading activation process of the type explained above should be reflected by a desynchronization of the alpha rhythm. And, indeed, as a variety of studies have shown, alpha desynchronization indicates a state of cortical activation (for recent reviews see, e.g., Pfurtscheller, 1992; Pfurtscheller & Klimesch, 1992; Pfurtscheller & Klimesch, 1991), whereas alpha synchronization reflects a state of cortical inactivity or "idling" (Pfurtscheller, 1992). Thus, alpha desynchronization might very well reflect the actual encoding process which is based on a frequency modulation in the relevant cell assemblies. The fact that during desynchronization a complex pattern of changes in EEG coherences can be found, as the interesting work of Petsche and his group demonstrates (e.g., Petsche and Rappelsberger, 1992) is well in line with the proposed interpretation.



31. In the identification of the relevant nodes or codes, feedback loops may play a decisive role. During the course of a search process, the activation status of the searched network is constantly transmitted back by means of feedback loops to a control system. The basic idea is that a control network converging in a particular control system is mapped onto the storage network. Consequently, the control system should be connected with the cortex by a dense network of axonal connections. Besides the basal ganglia, the thalamus with its thalamo-cortical projections to virtually all different cortical regions (e.g., Hoehl-Abrahao & Creutzfeldt, 1991) is one of those brain structures that fulfills this requirement. It is important to see that as compared to the thalamo-cortical network, the cortical network is orders of magnitude denser. Thus, each feedback loop serves a relatively large cortical field which also will be termed "alpha field".

32. To initiate a LTM search requires some vague information of where in the storage network to look for the relevant information. Because it would be highly inefficient to search the entire network once a search process is initiated, it is necessary to delimit the search area. In a theoretical sense, retrieval cues that give a rough description or some details of the relevant information can serve this purpose. In an anatomical sense, the thalamus might be a good candidate for delimiting the search area in the neocortex because the thalamo-cortical network may allow direct access to certain parts of the neocortex. Based on the current context of the WMS, retrieval cues are provided that enable the thalamus to activate particular thalamo-cortical pathways which start the search process in the neocortex. Thereby, specific and unspecific thalamic projections might provide access to specific sensoric or more abstract information, respectively. However, it should also be noted that in contrast to the traditional belief, the thalamus shows a much more complex pattern of different types of projections (e.g., Steriade et al., 1990, p. 40).

33. The thalamus and hippocampus probably are involved in quite different functions. Whereas the thalamus might serve as a relay station for searching and retrieving pure LTM information, the hippocampus (as well as other parts of the limbic system) might be important for the encoding and retrieval of concomitant episodic information. Note, however, that any search and retrieval process is embedded within a particular autobiographic context which defines the particular episodic meaning of that information which is retrieved from LTM. Thus, it must be expected that the functions of the hippocampus and the thalamus are closely interrelated.


34. Some researchers have suggested that the EEG frequency within the alpha band (of about 8-13 Hz) stems from the thalamus and induces synchronized neuronal activity in the cortex (Andersen & Andersson, 1968). In terms of EEG frequencies, the ideas described so far can be summarized as follows. A search processes in LTM starts if thalamo-cortical pathways are selectively activated, which means that a particular subset of these pathways shift their resting frequency within the alpha band. This shift in frequency serves as input frequency f to the cortical storage network where the search process spreads between different access points. In response to the search process, different cell assemblies start to oscillate with different frequencies f' within the alpha, beta and gamma bands. The status of the search process is constantly fed back by each activated cortical field (alpha field) that is served by a thalamo-cortical feedback loop oscillating within the range of alpha frequency. The higher cortical frequency f' is, the stronger is the concomitant increase in alpha frequency. The relevant information can be read out via those feedback loops responding with the highest frequency. Because we have assumed that alpha oscillations reflect the information processing in thalamocortical feedback loops, whereas gamma oscillations reflect pure cortical processing, a specific transition or interface between alpha and gamma oscillations must be postulated. The following two facts support this idea. First, Steriade et al. (1990, p. 147) report that the gamma rhythm (the 40 Hz rhythm in particular) is driven by neurons located in that cortical layer, which receives thalamic afferents. They conclude that thalamic input to the cortex serves as a trigger for rhythmic activation of specific cortical columns. Second, Pfurtscheller et al. (1994) observed a reciprocal relationship between alpha and the 40 Hz rhythm: If the 40 Hz rhythm synchronizes, alpha desynchronizes and vice versa.



35. If we proceed from the idea that memory codes are retrieved via longitudinal pathways linking thalamic nuclei with the cortex, and that alpha is the predominant rhythm reflecting the activity of these pathways, we arrive at the hypothesis that alpha frequency should be related to memory performance (see section I). We have tested this hypothesis (Klimesch, Schimke, Ladurner & Pfurtscheller, 1990; Klimesch, Schimke & Pfurtscheller, 1993) and found that - as compared to bad memory performers - good performers show a significantly higher alpha frequency. This result was found even in a resting state where subjects relaxed with eyes closed, but was most pronounced during actual retrieval attempts (Klimesch et al., 1990, Experiments 1 and 2). Furthermore, it could be demonstrated that attentional demands or interindividually different responses to increasing memory load (as an indicator of STM-span) are not responsible for the higher alpha frequency of good memory performers (Klimesch et al., 1993).

36. In a recently performed study, Klimesch, Schimke and Schwaiger (1994) were able to demonstrate that alpha power selectively responds to semantic LTM demands, whereas theta power selectively responds to episodic STM demands. In this study, an experimental design was used, that already proved useful in distinguishing semantic LTM from episodic STM (Kroll & Klimesch, 1992). The experimental design consisted of two parts. Subjects first performed a semantic congruency task in which they had to judge whether or not the sequentially presented words of concept-feature pairs (such as "eagle-claws" or "pea-huge") were semantically congruent. Then, without prior warning, they were asked to perform an episodic recognition task. This was done in an attempt to prevent subjects from using semantic encoding strategies and thus to increase episodic memory demands. In the episodic task, the same concept-feature pairs were used as targets and were presented together with distractor pairs (generated by repairing known concept-feature pairs). Now subjects had to judge whether or not a particular concept-feature pair was already presented during the semantic task. The results of Kroll and Klimesch (1992) indicated that semantic features speeded up semantic, but slowed down episodic decision times (for an extensive review on this topic see also Klimesch, 1994). With respect to the purpose of the present study, this result indicates that semantic and episodic memory processes can effectively be differentiated by using the design underlying Experiment 4 in Kroll and Klimesch (1992). According to the proposed hypothesis, it is expected that only in the semantic task should the most pronounced desynchronization (decrease in alpha band power) be observed in the alpha band. Because pairs of items are presented and because a subject can only perform the episodic and semantic task after the second item of a pair (i.e., the feature) is presented, a decrease in alpha band power as a response to increasing semantic task demands is expected only for the time period following the presentation of the feature. It is important to keep in mind that alpha power is well known to decrease (desynchronize) with increasing task difficulty and attentional demands. From the results found in Kroll and Klimesch (1992) we know that the episodic task is much more difficult than the semantic task. Thus, if task difficulty would be the only factor which is reflected by a decrease in alpha power, we would expect the most pronounced response to be observed during the presentation of the feature in the episodic task. This, however, was not the case. In support of our hypothesis it was found that alpha desynchronizes during the presentation of the feature in the semantic task. In the episodic task, on the other hand, theta power increased (synchronized) during the processing of the feature.


37. Since Scoville and Milner (1957) reported a severe anterograde amnesia for patient H.M. who had undergone a bilateral temporal lobectomy, including the hippocampal formation, and since Green and Arduini (1954) have found a dominant rhythmic electrical activity within the theta band in the hippocampus of rats, it has become obvious that theta activity of the hippocampus might be related to the encoding and/or retrieval of new information. Positive evidence came from studies which have documented that there is a preference for long-term potentiation (LTP) to occur in the hippocampal formation, and that theta activity induces or at least enhances LTP (e.g., Larson, Wong & Lynch, 1986; and Greenstein, Pavlides & Winson, 1988). The fact that LTP is considered the most important electrophysiological correlate for encoding new information, underlines the potential importance of hippocampal theta for memory processes in the WMS.

38. In trying to explain the possible functional significance of the theta rhythm in the human EEG, we assume that synchronized bursts of a small set of hippocampal pyramidal cells induce theta activity in selected but distributed cortical regions which are relevant for performing a particular task. Empirical findings support this view and indicate that theta band power increases with increasing (episodic) task demands (Klimesch, Schimke & Schwaiger, 1994). Research in animals also indicates that during behavioral activity, theta power increases (e.g., the review in Lopes da Silva, 1992).

39. One of the first questions that may arise when considering the proposed hypothesis is, why - in contrast to animals - theta is not a dominant rhythm in the human EEG. In an attempt to answer this question we first proceed from a theoretical consideration that is similar to the mechanisms that were proposed for accessing and retrieving LTM codes. It is assumed that hippocampo-cortical feedback loops induce synchronized rhythmic theta activity onto different regions of the neocortex where (e.g., by means of LTP) new information is encoded or freshly encoded information is retrieved. Given the basic assumption that new information always will be "added" or "attached" to related but already encoded information, only a small subset of the hippocampo-cortical feedback loops which are related to relevant cortical areas will be needed and thus will actually show synchronized theta activity. Because the human cortex is much larger than those in lower mammals and, as a consequence, holds much more LTM-information, the encoding of new information is a much more distributed process than in animals. Thus, if the percentage of synchronized hippocampo-cortical feedback loops is related to the size of the cortex (and to the hippocampus too, which is relatively much smaller in humans), this percentage will be orders of magnitudes smaller for humans as compared to animals.

40. Evidence for the view that only a small percentage of hippocampo-cortical feedback loops is synchronized comes from a re-examination of the pacemaker role of the septum in the production of the hippocampal theta rhythm (Petsche, Stumpf & Gogolak, 1962; Stewart & Fox, 1990). In addition to cholinergic projections, a large fraction of the septo-hippocampal projections terminate on inhibitory (GABA-ergic) hippocampal interneurons (Freund & Antal, 1988; see also the reviews in Lopes da Silva, 1992; and Stewart & Fox, 1990). Based on these and related findings, Stewart and Fox (1990) assume that the septal input might organize the hippocampal theta activity via rhythmic inhibition of hippocampal interneurons. This view is in agreement with the fact that hippocampal interneurons are more likely to behave as theta cells (Fox & Ranck, 1981) than burst firing pyramidal neurons. In agreement with this fact, a simulation model (Traub, Miles & Wong, 1989) reveals that in contrast to interneurons, only a small percentage of the pyramidal cells display synchrony.

41. With respect to the question, whether theta activity can be observed in the EEG, these findings which were obtained from microelectrodes in the hippocampus are of outstanding importance. Biophysically, theta frequency in the hippocampus, deep inside the brain, would be difficult to detect from scalp electrodes. The crucial condition to detect theta as a dominant rhythm in the EEG would be that most of the burst firing hippocampal pyramidal cells that project to other parts of the cortex would fire in synchrony. However, as we have already noted, according to Leung (1980), Traub et al. (1989) and Lopes da Silva (1992), this is not the case. And indeed, as judged by visual inspection but in contrast to spectral analysis, theta activity usually is absent in the EEG of normal, wake adults.

42. The fact that only a small percentage of the pyramidal neurons displays synchrony, agrees with the idea that hippocampo-cortical feedback loops induce synchronous theta activity into selected cortical areas where new information is encoded or fresh information is retrieved. This is type 2 or selective synchronization that means activation. Given the fact that theta frequency induces or at least enhances LTP (see also Lopes da Silva, 1992), it seems tempting to assume that theta activity, induced into selected cortical areas, reflects a process to encode or retrieve new information by keeping or putting selected cortical areas into a state of resonance. This assumption comes very close to a theory of resonant phase-locked hippocampo-cortical loops, proposed by Miller (1991).


43. A possible objection against our central hypothesis that brain oscillations are the basic phenomenon of cortical information processing may come from researchers using event related potentials (ERPs) to study cognitive and memory processes in particular. They may argue that late components of ERPs reflect completely different types of cortical processes such as, for example, time locked threshold changes that regulate the degree of excitability in neuronal networks (e.g., Birbaumer & Elbert, 1988; Elbert, 1992). Given the fact that almost all of the memory studies in cognitive electrophysiology focus on late components of ERPs, this objection could seriously threaten the general validity of our hypothesis.

44. However, it may also be argued that late components of ERPs are the result of synchronous oscillations that are transiently phase locked in response to a relevant event or stimulus. Summed up over several trials, a waveform (the ERP) would be generated that shows the typical succession of positive and negative "peaks" (or ERP components). If we proceed from this idea, it becomes evident that only (or at least primarily) those types of oscillations would be capable of generating late ERP components that indeed respond with coupled type 2 synchronization to an increase in respective task demands. Note that alpha and beta tend to desynchronize with increasing task demands. Hypothetically, there are only two possible candidates (see section II.3; paragraph 15): theta and gamma frequency. Because gamma frequency is much too high and its amplitudes much too small to generate a typical ERP, theta frequency remains the most plausible candidate.

45. This proposal, of course, does not mean that theta is the only generator for ERPs. In addition, there may be other processes such as very slow oscillations in the delta band (below 4 Hz) and/or threshold changes in large parts of the cortical network that also may have a strong influence on the waveform of ERPs. Weak influences may even be due to type 2 synchronizations within the alpha band.

46. The most obvious ERP component that might reflect phase locked evoked theta activity is the P300 for the following two reasons: First, because of the (typical) latency and form of the P300, this ERP component shows the most significant power in the theta band as frequency decompositions indicate (Basar & Stampfer, 1985). Second, the (typical) functional meaning of the P300, particularly the process of "updating" (Donchin & Coles, 1988), is well related to central functions of the WMS.

47. Now, let us consider the most speculative part of our proposal which links phase locked theta activity to the P300 component of event related potentials (ERPs). Keeping in mind that a single cycle of the theta rhythm consists of an inhibitory and a disinhibited or excitatory phase and that only in the latter, bursts of action potentials are sent to selected cortical areas, the question arises: With which phase of the cycle does task related (episodic) processing start? Does it start with the inhibitory or the excitatory phase? In referring to the argument that theta is induced in selected parts of the cortical network, it seems plausible to assume that episodic processing starts with the inhibitory phase in order to maximize the impact of the distributed activation of the relevant parts of the network by reducing irrelevant background activity through the inhibitory phase. As a result of this assumption, and because only a small percentage of the burst firing pyramidal cells are synchronized through the excitatory phase, the outcome should be a positive going deflection in the EEG, time locked to the presentation of an adequate stimulus.

48. If this is true, it should be possible to record evoked theta activity from the scalp in response to an appropriate stimulus or event. Based on theoretical considerations and experimental evidence (e.g., Leung, 1980), Lopes da Silva (1992, p. 93) concludes that appropriate stimuli or events induce evoked responses that depend on the phase of theta frequency. Therefore, evoked theta activity may be viewed as synchronized phase locked and thus amplified theta frequency which occurs in response to an appropriate event. The issue of interest is whether evoked or event-related theta activity can be detected as a response to increased episodic memory demands.

49. We have emphasized that one of the most important functions of the WMS that we have related to hippocampal information processing is the encoding of contextual or episodic information. Thus, if the P300 really stems from phase-locked hippocampal theta activity, the (typical) functional meaning of the P300 should be related to the encoding of contextual and the encoding of new information.

50. There is some evidence for this view. Donchin's updating hypothesis (e.g., Donchin & Coles, 1988) is one of the best examples. It is well established that the P300 amplitude is related to the degree of contextual encoding (e.g., Donchin & Coles, 1988), expectancy (or subjective probability), and the amount of effort which is also reflected by the amount of information transmitted to a subject (see e.g., Johnson's triarchic model in Johnson, 1986; and the summary in Verleger, 1988, p. 351). It is important to note that Verleger (1988), who is challenging Donchin's updating hypothesis, is not challenging the significance of the P300 with respect to contextual encoding. His argument basically is that the P300 does not reflect the "updating" but instead the "closure" of expectancies.

51. A positive relationship between the P300 and the consolidation of memory codes (as a typical hippocampal function) was demonstrated by some of those studies reporting a Dm-effect. Several studies have shown that ERPs recorded during the encoding of words (or pictures) that were later remembered were more positive than ERPs to words (or pictures) that were not remembered (Sanquist et al., 1980; Karis et al., 1982, 1984; Johnson, et al., 1985; Neville et al., 1986; Fabiani et al., 1986; Paller et al., 1987; Fabiani et al., 1990; Friedman, 1990a, b; and see also the indirect evidence provided by e.g., Noldy et al., 1990; and the review in Paller, 1993). This difference in ERPs during encoding which was found within the region of the typical P300 or a late positive component was termed "Dm" (for Difference based on later memory performance; Paller et al., 1987) or Dm-effect. When reviewing this research it is interesting to see that particularly the P300 does not reflect the processing of semantic information (i.e., the encoding of a stimulus per se) but instead the processing of episodic information. Results reported by Karis et al. (1984) and Fabiani et al. (1986, 1990) are in good agreement with this interpretation. They presented subjects with different series of words which had to be recalled immediately after a list was presented and found that words later recalled elicited larger P300s than words not recalled. In addition, Fabiani et al. (1990) were able to demonstrate that this relationship between the P300 amplitude and episodic memory performance holds only if subjects use rote learning (which is based on the encoding of contextual and thus episodic information) but not if subjects use semantic encoding strategies (such as organizing the words into meaningful sentences). Thus, the Dm (with respect to the P300) most likely reflects episodic encoding processes and as a result, this type of Dm-effect which is based on the P300 component becomes the weaker; the more semantic encoding processes predominate.

52. Electrophysiological recordings with electrodes implanted in the hippocampus have not provided clear evidence for the view that the P300 is generated in the hippocampus (e.g., Polich & Squire, 1993; and the literature reviewed in this article). Because theta is generated in the septum and because other parts of the limbic system also exhibit theta frequency, the crucial question is, whether or not it can be demonstrated that theta activity (or the P300) varies as a function of (episodic) memory performance and that at the same time the hippocampus is involved in the modification of theta activity (or the P300).

53. An interesting study by Smith and Halgren (1989) who focused on the word repetition effect in a recognition task provided evidence for this view. It is well known that old words (repeated words) elicit a larger P300 than new words (e.g., Sanquist et al., 1980; and Johnson et al., 1985). Smith and Halgren (1989) repeated the targets in the recognition task in each of a set of nine blocks of 20 words (consisting of 10 targets and 10 new words) and found that the amplitude difference between repeated and new words did not change with the number of repetitions (i.e., the number of blocks). Recognition performance, of course, increased with the number of blocks, but this increase in performance was not reflected by the amplitude differences between the repeated and not repeated words which remained constant with the number of repetitions. Because these results were found for normal subjects as well as for patients with unilateral (left or right) anterior temporal lobectomy, it was concluded that the hippocampus is not involved in the increase of recognition performance over different blocks, which can be explained as an increase in implicit memory performance. Most important, however, the baseline recognition performance was significantly lower for the patients with a left temporal lobectomy who from the very beginning also failed to show a significant P300 amplitude difference between new and old words. This latter result is consistent with the hypothesis that the hippocampus (in the dominant left hemisphere) might be capable of modifying a P300 that reflects explicit, episodic memory performance.


54. The main purpose of this article is to encourage an integrative and interdisciplinary view on memory processes. As a result of this attempt, new experiments can be performed that will be capable of critically evaluating the proposed hypotheses. A promising empirical approach would be to analyze event-related shifts in EEG power within the theta and alpha bands in amnesic subjects who perform different types of memory tasks.

55. If it is true that oscillations are the mandatory basis for information transmission in the cortex and possibly in the entire brain, a better understanding of the nature of oscillations would be essential for an integrative view in cognitive neuroscience. For cognitive psychology this finally would mean to describe cognitive processes in terms of oscillations and for cognitive psychophysiology this would mean to focus primarily on the analysis of certain, carefully selected frequency bands in addition to the study of event-related potentials.


This research was supported by the Austrian "Fonds zur Foerderung der wissenschaftlichen Forschung", Project S-4904 and Project P-10235.

I wish to thank Stevan Harnad and anonymous reviewers for their helpful suggestions. In particular, I am grateful for the insightful critical comments of Niels Birbaumer on an earlier draft of this manuscript.


Andersen, P. & Andersson, A.A. (1968). Physiological basis of the alpha rhythm. New York: Century Crofts.

Anderson, J.R. (1983). A spreading activation theory of memory. Journal of Verbal Learning and Verbal Behavior, 22, 261-295.

Arnolds, D., Lopes da Silva, F., Aitink, J., Kamp, A. & Boeijinga, P. (1980). In G. Pfurtscheller & Lopes da Silva (Eds.), Rhythmic EEG activities and cortical functioning (pp. 91-102). Amsterdam: Elsevier.

Baddeley, A. (1992). Working Memory. Science, 255, 5044, 556-559.

Baddeley, A. (1984). Neuropsychological evidence and the semantic/ episodic distinction (Commentary to Tulving). The Behavioral and Brain Sciences, 7, 238-239.

Basar, E. & Bullock, T.H. (Eds.). (1992). Induced rhythms in the brain. Boston: Birkhaeuser.

Basar, E. & Stampfer, H.G. (1985). Important associations among EEG-dynamics, event-related potentials, short-term memory and learning. International Journal of Neuroscience, 26 (3-4), 161-180.

Berger, H. (1929). Ueber das Elektroenkephalogramm des Menschen. Archiv fuer Psychiatrie und Nervenkrankheiten, 87, 527-570.

Birbaumer, N. & Elbert, Th. (1988). P3: Byproduct of a byproduct. Behavioral and Brain Sciences, 11(3), 375-377.

Braitenberg, V. & Schuez, A. (1991). Anatomy of the cortex. New York: Springer Verlag.

Carpenter, G.A. & Grossberg, S. (1993). Normal and amnesic learning, recognition and memory by a neuronal model of cortico-hippocampal interactions. TINS, 16, 131-137.

Donchin, E. & Coles, M.G.H. (1988). Is the P300 component a manifestation of context updating? Behavioral and Brain Sciences, 11, 357-374.

Elbert, T. (1992). A theoretical approach to the late components of the event-related brain potential. In A. Aertsen & V. Braitenberg (Eds.). Information processing in the cortex (225-245). Berlin: Springer Verlag.

Engel, A.K., Koenig, P., Kreiter, A., Schillen, Th. & Singer, W. (1992). Temporal coding in the visual cortex: new vistas on integration in the nervous system. TINS, 15(6), 218-226.

Fabiani, M., Karis, D. & Donchin, E. (1986). P300 and recall in an incidental memory paradigm. Psychophysiology, 23 (3), 298-308.

Fabiani, M., Karis, D. & Donchin, E. (1990). Effects of mnemonic strategy manipulation in a Von Restorff paradigm. Electroencephalography and clinical Neurophysiology, 75, 22-35.

Fox, S.E. & Ranck, J.B. Jr. (1981). Electrophysiological characteristics of hippocampal complex-spike cells and theta cells. Experimental Brain Research, 41, 399-410.

Freund, T. & Antal, M. (1988). Hippocampal interneurons. Nature, 336, 170-173.

Friedman, D. (1990a). Cognitive event-related potential components during continuous recognition memory for pictures. Psychophysiology, 27, 136-148.

Friedman, D. (1990b). ERPs during continuous recognition memory for words. Biological Psychology, 30, 51-87.

Gray, Ch.M., Koenig, P., Engel, A.K. & Singer, W. (1989). Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature, 338, 335-337.

Gray, C. & Singer, W. (1987). Stimulus-dependent neuronal oscillations in the cat visual cortex area 17. Neuroscience (Supplement), 22, 434

Green, J.D. & Arduini, A.A. (1954). Hippocampal electrical activity in arousal. Journal of Neurophysiology, 17, 533-547.

Greenstein, Y.J., Pavlides, C. & Winson, J. (1988). Long-term potentiation in the dentate gyrus is preferentially induced at theta rhythm periodicity. Brain Research, 438, 331-334.

Grossberg, S. (1980). How Does a Brain Build a Cognitive Code? Psychological Review, 87, 1-51.

Grossberg, S. & Stone, G. (1986). Neuronal Dynamics of Word Recognition and Recall: Attentional Priming, Learning, and Resonance. Psychological Review, 93, 46-74.

Hebb, D.O. (1949). The first stage of perception. In D.O. Hebb (Ed.), The Organization of behavior (pp. 60-78). New York: Wiley.

Hoehl-Abrahao, J.C. & Creutzfeldt, O.D. (1991). Topographical mapping of the thalamocortical projections in rodents and comparison with that in primates. Experimental Brain Research, 87, 283-294.

Johnson, R.Jr. (1986). A triarchic model of P300 amplitude. Psychophysiology, 23, 367-384.

Johnson, R., Pfefferbaum, A. & Kopell, B.S. (1985). P300 and long-term memory: Latency predicts recognition performance. Psychophysiology, 22 (5), 497-507.

Karis, D., Bashore, T., Fabiani, M., & Donchin, E. (1982). P300 and memory. Psychophysiology, 19, 328.

Karis, D., Fabiani, M. & Donchin, E. (1984). "P300" and memory: Individual differences in the von Restorff effect. Cognitive Psychology, 16, 177-216.

Klimesch, W. (1987). A connectivity model for semantic processing. Psychological Research, 49, 53-61.

Klimesch, W. (1994). The structure of long-term memory: A connectivity model of semantic processing. Hillsdale, NJ, Lawrence Erlbaum.

Klimesch, W., Schimke, H. & Ladurner, G. (1988). Die Suchzeit fuer episodische und semantische Information. Sprache & Kognition, 7, 129-143.

Klimesch, W., Schimke, H., Ladurner, G. & Pfurtscheller, G. (1990). Alpha frequency and memory performance. Journal of Psychophysiology, 4, 381-390.

Klimesch, W., Schimke, H. & Pfurtscheller, G. (1993). Alpha frequency, cognitive load and memory performance. Brain Topography, 5, 1-11.

Klimesch, W., Schimke, H., Doppelmayr, M., Ripper, B., Schwaiger, J. & Pfurtscheller, G. (1994). Event-related desynchronization (ERD) and the Dm-effect: Does alpha desynchronization during encoding predict later recall performance? Submitted paper.

Klimesch, W., Schimke, H. & Schwaiger, J. (1994). Episodic and semantic memory: An analysis in the theta and alpha band. Electroencephalography and clinical Neurophysiology, in press.

Koestner, J. (1985). Functional consequences of passive membrane properties of the neuron. In E.R. Kandel & J.H. Schwartz (Eds.), Principles of neuronal science (pp. 66-74). New York: Elsevier.

Kroll, N.E.A. & Klimesch, W. (1992). Semantic memory. Complexity or connectivity? Memory & Cognition, 20, 192- 210.

Larson, J., Wong, D. & Lynch, G. (1986). Patterned stimulation at the theta frequency is optimal for the induction of hippocampal long-term potentiation. Brain Research, 368, 347-350.

Lashley, K.S. (1950). In Search of the Engram. Society for Experimental Biology, Symposium, No. 4, 454-482.

Leung, L. (1980). Behavior-dependent evoked potentials in the hippocampal CA1 region of the rat. I. Correlation with behavior and EEG., Brain Research, 198, 95-117.

Lopes da Silva, F.H. (1992). The rhythmic slow activity (theta) of the limbic cortex: An oscillation in search of a function. In E. Basar & T.H. Bullock (Eds.), Induced rhythms in the brain (pp. 83-102). Boston: Birkhaeuser.

Lopes da Silva, F.H., Witter, M.P., Boeijinga, P.H. & Lohman, A.H.M. (1990). Anatomic organization and physiology of the limbic cortex. Physiological Reviews, 70, 453-511.

Markowitsch, H.J. (1983). Transient global amnesia. Neuroscience and Biobehavioral Reviews, 7, 35-43.

Markowitsch, H.J. (1984). Can amnesia be caused by damage of a single brain structure? Cortex, 20, 27-45.

Markowitsch, H.J. & Pritzel, M. (1985). The neuropathology of amnesia. Progress in Neurobiology, 25, 189-288.

Miller, R. (1991). Cortico-hippocampal interplay and the representation of contexts in the brain. Berlin: Springer Verlag.

Neville, H.J., Kutas, M., Chesney, G. & Schmidt, A.L. (1986). Event-related brain potentials during initial encoding and recognition memory of congruous and incongruous words. Journal of Memory and Language, 25, 75-92.

Noldy, N.E., Stelmack, R.M. & Campbell, K.B. (1990). Event-related potentials and recognition memory for pictures and words: The effects of intentional and incidental learning. Psychophysiology , 27 (4), 417-428.

Paller, K.A. (1993). Elektrophysiologische Studien zum menschlichen Gedachtnis. EEG & EMG, 24, 24-33.

Paller, K.A., Kutas, M. & Mayes, A.R. (1987). Neuronal correlates of encoding in an incidental learning paradigm. Electroencephalography and clinical Neurophysiology, 67, 360-371.

Petsche, H. & Rappelsberger, P. (1992). Is there any message hidden in the human EEG? In E. Basar, & T.H. Bullock (Eds.), Induced rhythms in the brain (pp. 103-116). Boston: Birkhaeuser.

Petsche, H., Stumpf, C. & Gogolak, G. (1962). The significance of the rabbit's septum as a relay station between the midbrain and the hippocampus. Electroencephalography and Clinical Neurophysiology, 19, 25-33.

Pfurtscheller, G. (1992). Event-related synchronization (ERS): an electrophysiological correlate of cortical areas at rest. Electroencephalography and clinical Neurophysiology, 83, 62-69.

Pfurtscheller, G. & Klimesch, W. (1991). Event-related desynchronization during motor behavior and visual information processing. In C.H.M. Brunia, G. Mulder, & M.N. Verbaten (Eds.), Event-related brain research (pp. 58-65). Amsterdam: Elsevier.

Pfurtscheller, G. & Klimesch, W. (1992). Functional topography during a visuoverbal judgment task studied with event-related desynchronization mapping. Journal of Clinical Neurophysiology, 9, 120-131.

Pfurtscheller, G., Flotzinger, D. & Neuper, Ch. (1994). Differentiation between finger, toe and tongue movement in man based on 40-Hz EEG. Electroencephalography and clinical Neurophysiology, 90, 456-460.

Polich, J. & Squire, L.R. (1993). P300 from amnesic patients with bilateral hippocampal lesions. Electroencephalography and clinical Neurophysiology, 86, 408-417.

Pulvermueller, F., Preissl, H., Eulitz, C., Pantev, Ch., Lutzenberger, W., Elbert, Th. & Birbaumer, N. (1994). Brain Rhythms, Cell Assemblies and Cognition: Evidence From the Processing of Words and Pseudowords. PSYCOLOQUY 5(48) brain-rhythms.1.pulvermueller.

Sanquist, Th.F., Rohrbaugh, J.W., Syndulko, K. & Lindsley, D.B. (1980). Electrocortical signs of levels of processing: Perceptual analysis and recognition memory. Psychophysiology, 17, 568-576.

Scoville, W.B. & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology, Neurosurgery and Psychiatry, 20, 11-21.

Smith, M.E. & Halgren, E. (1989). Dissociation of Recognition Memory Components Following Temporal Lobe Lesions. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 50-60.

Squire, L.R. (1992). Memory and the hippocampus: A synthesis from findings with rats, monkeys, and humans. Psychological Review, 99, 195-231.

Squire, L.R., Knowlton, B. & Musen, G. (1993). The structure and organization of memory. Annual Review of Psychology, 44, 453-495.

Steriade, M., Jones, E. & Llinas, R. (1990). Thalamic oscillations and signaling. New York: John Wiley.

Stewart, M. & Fox, S.E. (1990). Do septal neurons pace the hippocampal theta rhythm? Trends in Neuroscience, 13, 163-168.

Strogatz, St. & Stewart, I. (1993). Coupled oscillators and biological synchronization. Scientific American, 269(6), 68-75.

Teyler, T.J. & DiScenna, P. (1986). The hippocampal memory indexing theory. Behavioral Neuroscience, 100, 147-154.

Traub, R., Miles, R. & Wong, R. (1989). Model of the origin of rhythmic population oscillations in the hippocampal slice. Science, 243, 1319-1325.

Tulving, E. (1984). Precis of elements of episodic memory. The Behavioral and Brain Sciences, 7, 223-268.

Vanderwolf, C. & Robinson, T. (1981). Retico-cortical activity and behavior: A critique of the arousal theory and a new synthesis. Behavioral and Brain Sciences, 4, 459-514.

Verleger, R. (1988). Event-related potentials and cognition: A critique of the context updating hypothesis and an alternative interpretation of P3. Behavioral and Brain Sciences, 11, 343-427.

Volume: 6 (next, prev) Issue: 06 (next, prev) Article: 1 (next prev first) Alternate versions: ASCII Summary