The 2004 EJN Lecture was an attempt to lay out further aspects of a developing neurobiological theory of hippocampal function [Morris, R.G.M., Moser, E.I., Riedel, G., Martin, S.J., Sandin, J., Day, M. & O'Carroll, C. (2003) Phil. Trans. R. Soc. Lond. B Biol. Sci., 358, 773–786.] These are that (i) activity-dependent synaptic plasticity plays a key role in the automatic encoding and initial storage of attended experience; (ii) the persistence of hippocampal synaptic potentiation over time can be influenced by other independent neural events happening closely in time, an idea with behavioural implications for memory; and (iii) that systems-level consolidation of memory traces within neocortex is guided both by hippocampal traces that have been subject to cellular consolidation and by the presence of organized schema in neocortex into which relevant newly encoded information might be stored. Hippocampal memory is associative and, to study it more effectively than with previous paradigms, a new learning task is described which is unusual in requiring the incidental encoding of flavour–place paired associates, with the readout of successful storage being successful recall of a place given the flavour with which it was paired. NMDA receptor-dependent synaptic plasticity is shown to be critical for the encoding and intermediate storage of memory traces in this task, while AMPA receptor-mediated fast synaptic transmission is necessary for memory retrieval. Typically, these rapidly encoded traces decay quite rapidly over time. Synaptic potentiation also decays rapidly, but can be rendered more persistent by a process of cellular consolidation in which synaptic tagging and capture play a key part in determining whether or not it will be persistent. Synaptic tags set at the time of an event, even many trivial events, can capture the products of the synthesis of plasticity proteins set in train by events before, during or even after an event to be remembered. Tag–protein interactions stabilize synaptic potentiation and, by implication, memory. The behavioural implications of tagging are explored. Finally, using a different protocol for flavour–place paired associate learning, it is shown that rats can develop a spatial schema which represents the relative locations of several different flavours of food hidden at places within a familiar space. This schema is learned gradually but, once acquired, enables new paired associates to be encoded and stored in one trial. Their incorporation into the schema prevents rapid forgetting and suggests that schema play a key and hitherto unappreciated role in systems-level memory consolidation. The elements of what may eventually mature into a more formal neurobiological theory of hippocampal memory are laid out as specific propositions with detailed conceptual discussion and reference to recent data.