The Implicit Association Test (IAT) is designed to measure implicit attitudes, and is often claimed to reveal prejudicial attitudes that are at odds with explicit attitudes. Numerous proposals as to the information processing mechanisms underlying human performance on the IAT have been suggested, many or all of which may well play a role. This makes it difficult to study them and their interactions experimentally in an efficient manner. We describe a localist connectionist model that simulates human performance on the IAT and that allows us to explore many of the proposed explanations, by comparing the results with observations from actual experiments with human subjects. By simulating the performance of virtual subjects, the model also makes it possible to conduct theoretical experiments that could not be undertaken with real subjects in the real world.