This article attempts an overview of the mechanism of NMDAR-dependent long-term potentiation (LTP) and its role in hippocampal networks. Efforts are made to integrate information, often in speculative ways, and to identify unresolved issues about the induction, expression and molecular storage processes. The pre/post debate about LTP expression has been particularly difficult to resolve. The following hypothesis attempts to reconcile the available physiological evidence as well as anatomical evidence that LTP increases synapse size. It is proposed that synapses are composed of a variable number of trans-synaptic modules, each having presynaptic release sites and a postsynaptic structure that can be AMPAfied by the addition of a hyperslot assembly that anchors 10-20 AMPA channels. According to a newly developed view of transmission, the quantal response is generated by AMPA channels near the site of vesicle release and so will depend on whether the module where release occurs has been AMPAfied. LTP expression may involve two structurally mediated processes: (i) the AMPAfication of existing modules by addition of hyperslot assemblies: this is a purely postsynaptic process and produces an increase in the probability of an AMPA response, with no change in the NMDA component; and (ii) the addition of new modules: this is a structurally coordinated pre/post process that leads to LTP-induced synapse enlargement and potentiation of the NMDA component owing to an increase in the number of release sites (the number of NMDA channels is assumed to be fixed). The protocol used for LTP induction appears to affect the proportion of these two processes; pairing protocols that involve low-frequency presynaptic stimulation induce only AMPAfication, making LTP purely postsynaptic, whereas high-frequency stimulation evokes both processes, giving rise to a presynaptic component. This model is capable of reconciling much of the seemingly contradictory evidence in the pre/post debate. The structural nature of the postulated changes is relevant to a second debate: whether a CaMKII switch or protein-dependent structural change is the molecular memory mechanism. A possible reconciliation is that a reversible CaMKII switch controls the construction of modules and hyperslot assemblies from newly synthesized proteins.