Much evidence indicates that the perirhinal cortex of the temporal lobe is involved in the familiarity discrimination aspect of recognition memory. All previously published models of familiarity discrimination in the perirhinal cortex are based on Hebbian learning. Here we present a biologically plausible model based on anti-Hebbian learning. When the responses of neurons providing input to the familiarity discrimination network are correlated (as is indicated by experimental data), then the anti-Hebbian model achieves a much higher capacity (up to thousands of times) and hence a crucially higher efficiency than models based on Hebbian learning.