A network model with some general properties of hippocampal area CA3, and the results of its simulation on a massively parallel processor, are described. This network performs the tasks of recent declarative memory including recovery of complete traces from partial cues and recognition of familiarity. Immediate recurrent inhibition is essential for providing sensitivity to small cues while preventing spurious recall. Tonic inhibition seems to set the retrieval speed/accuracy trade-off. Delayed inhibition resets hippocampal activity. The behavior under excessive or deficient inhibition resembles that of amnesics with lesions in brainstem areas known to modulate hippocampal inhibition. The rate of recall and dynamics of inhibition by the model are similar to those inferred to occur in the human hippocampus from unit and evoked potential recordings. The model suggests a mechanism whereby the hippocampus can control its own plasticity. These simulations demonstrate that the retrieval mechanism in several hippocampal models is feasible and that the theta rhythm and the cognitive evoked potentials may be generated by synaptic events modulating network parameters.