This paper reviews some of the recent research from our laboratory in which event-related brain potentials (ERPs) are used to study processes engaged by variants of tests of recognition and cued recall. The research has been guided by the framework developed over the past few years by Jacoby and his colleagues, and its results speak to several issues relevant to that framework.

Recent introductions to cognitive ERPs can be found in Picton et al. (1995); Rugg and Coles (1995a) and Rugg and Coles (1995b) also provide a detailed discussion of the assumptions underlying the use of ERPs in cognitive psychology. Briefly, ERPs represent scalp-recorded changes in the brain’s electrical activity (the electroencephalogram or EEG) time-locked to some definable event such as the presentation of a word. The magnitude of these changes is typically small in comparison to the amplitude of the ‘background’ EEG, which is in effect the noise from which the ERP ‘signal’ has to be extracted. ERP waveforms with satisfactory signal-to-noise ratios are obtained by averaging the EEG samples obtained on a number of trials (typically, between 20 and 50) belonging to the same experimental condition. The averaged waveforms represent estimates of time-locked neural activity elicited by the presentation of stimuli belonging to different experimental conditions.

There are several reasons why ERPs are useful for studying cognitive function in general, and memory in particular. First, neural activity (more accurately, that fraction detectable at the scalp) associated with the processing of different classes of stimuli can be measured with a temporal resolution which is sufficient to detect the neural correlates of cognitive processes that may be active for only a few tens of milliseconds. Thus, upper-bound estimates of the time required by the nervous system to discriminate between different classes of stimulus (e.g. old and new words in a recognition memory test) can be made directly. This level of temporal resolution is presently unattainable with other functional neuroimaging techniques, such as positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), that depend on haemodynamic measures. A second benefit of the ERP technique, central to the research reviewed below, is that ERP waveforms can be formed ‘off-line’, after the experimental trials have been sorted into different conditions on the basis of an analysis of the subject’s behaviour. Thus, again in contrast to PET and fMRI, it is easy to obtain and compare records of brain activity associated with different classes of response to the same experimental items (e.g. hits vs. misses, or false alarms vs. correct rejections).

Finally, ERPs can be used to investigate whether different experimental conditions engage functionally dissociable cognitive processes. This application of the technique necessitates the recording of ERP waveforms from sufficient scalp sites to permit the spatial configurations of the electrical field associated with the different experimental conditions to be characterised. Its rationale rests on the assumption that if two experimental conditions are associated with qualitatively different patterns of scalp electrical activity, then there are strong grounds for proposing that the conditions engaged at least partially non-overlapping neural, and hence functional, processes. By contrast, purely quantitative differences in the ERPs associated with two conditions constitute evidence for different levels of engagement of the same neural/functional processes, at least so far as can be detected from scalp recordings. In the research we describe below, the analysis of the spatial distribution (the ‘scalp topography’) of ERP effects associated with different experimental conditions occupies a central place; when two topographies differ, we take this as highly suggestive that the cognitive processes engaged in the respective experimental conditions are at least partially functionally dissociable.

It is important to note that the logic of using differences in scalp topography to identify functional dissociations does not require that the locations of the intracerebral generators of the ERP effects in question be known. This is fortunate, as identifying the brain regions responsible for generating or modulating cognitive ERPs, while highly desirable, is formidably difficult. In this respect, the ERP technique compares very poorly with PET and fMRI, which allow task-related changes in cerebral activity to be localised with a spatial resolution of less than a centimetre. At the present time, methods for the formal modeling of the diffusely distributed, long latency ERP effects that are typically found in studies of memory are poorly developed, and have yet to be widely applied. Thus in the course of our review we do no more than offer conjectures about the possible brain origins of the ERP effects we describe, drawing when possible on evidence from relevant neuropsychological and functional neuroimaging studies.