Over the last half century, the experimental study of human memory has departed from the earlier concept of a unitary faculty, with the increase in knowledge leading to differentiation between subsystems of memory, often based on the study of neuropsychological patients. Although foreshadowed by the classic work of William James (1890), the current approach to the fractionation of memory probably began with Hebb's (1949) proposal of a distinction between short-term memory (STM), based on temporary electrical activity within the brain, and longterm memory (LTM), based on the development of more permanent neurochemical changes. He even proposed a learning mechanism, a concept that continues to be influential in neurobiological theorizing (see Burgess et al. 2001). Experimental evidence for a distinction between STM and LTM began to appear a decade later with the demonstration by Brown (1958) and Peterson & Peterson (1959) of the rapid forgetting of small amounts of material when ongoing rehearsal was prevented. They proposed that this forgetting reflected the decay of a short-term trace, a process they distinguished from long-term forgetting, which was attributed to interference among longterm memory representations. This view was resisted, with the counter claim made that all forgetting could be interpreted within a single stimulus-response association framework (Melton 1963). The question of whether shortterm forgetting reflects trace decay or interference remains unresolved (Cowan et al. 2000; Service 1998). During the 1960s, however, experimental evidence from a range of sources seemed to point increasingly strongly to the need to distinguish between STM and LTM on grounds other than type of forgetting. Neuropsychological evidence was particularly influential, with patients suffering from the classic amnesic syndrome showing grossly impaired LTM, coupled with total preservation of performance on a range of tasks associated with STM (Baddeley & Warrington 1970). Anatomically, the amnesic syndrome has most strongly been associated with damage to the hippocampus (Milner 1966), although it could result from damage to a series of structures that broadly make up the Papez circuit (see Aggleton & Pearce 2001). The STM-LTM distinction was further supported by patients showing the opposite dissociation, with STM performance impaired and LTM preserved (Shallice & Warrington 1970). By the late 1960s, a range of two-component models was being proposed, of which the most influential was that of Atkinson & Shiffrin (1968). In this model, information was assumed to come in from the environment, be processed by a short-term storage system and then fed into LTM. Probability of learning was assumed to depend on time held within the short-term store. STM was also