The symptoms of schizophrenia may be associated with reductions in NMDA receptor (NMDAR) function. This is suggested by the psychotomimetic effects of NMDA antagonists, the ameliorative effects of NMDAR indirect agonists, elevated levels of the NMDA antagonist N-acetyl-aspartyl-glutamate (NAAG) in schizophrenic brain, and findings from recent genetic studies. However, the link between reduced NMDAR function and the behavioral features of schizophrenics has not been made explicit. Here we present a network simulation of hippocampal function, focused on retrieval of verbal stimuli in human memory tasks. Specifically, we trained a computational model of the hippocampal complex to perform a context-dependent paired associate task, a free recall task with category clustering, and the transitive inference (TI) task. In this network, direct perforant pathway input from entorhinal cortex to region CA1 provides the basis for semantic context cueing during initial encoding and retrieval, allowing selective retrieval on the basis of category cues. Alterations in the magnitude of this direct perforant pathway input to region CA1 causes impairments in use of organizational strategies for memory, accounting for specific features of memory dysfunction in schizophrenics and in normals treated with ketamine. This model provides a theoretical link between cellular physiological changes and specific cognitive symptoms. As such, it can shed light on the etiology of schizophrenia in a fundamental way, and also holds the promise of pointing the way to more effective treatments.