Paired-associate learning is often used to examine episodic memory in humans (1). Animal models include the recall of foodcache locations by scrub jays (2) and sequential memory (3,4). Here we report a model in which rats encode, during successive sample trials, two paired associates (flavours of food and their spatial locations) and display better-than-chance recall of one item when cued by the other. In a first study, pairings of a particular foodstuff and its location were never repeated, so ensuring unique ‘what–where’ attributes. Blocking N-methyl-D-aspartate receptors in the hippocampus—crucial for the induction of certain forms of activity-dependent synaptic plasticity (5,6)— impaired memory encoding but had no effect on recall. Inactivating hippocampal neural activity by blocking a-amino-3- hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) receptors impaired both encoding and recall. In a second study, two paired associates were trained repeatedly over 8 weeks in new pairs, but blocking of hippocampal AMPA receptors did not affect their recall. Thus we conclude that unique what–where paired associates depend on encoding and retrieval within a hippocampal memory space (7,8) with consolidation of the memory traces representing repeated paired associates in circuits elsewhere.