WE tested the hypothesis that the fine form selectivity of anterior inferotemporal (AIT) neurones represents long-term memory of object forms. Prior to neuronal recording, two monkeys were trained to learn a set of computer-generated Fourier descriptors (FDs). A slight alteration of one FD parameter produces a form with similar global features. We found that these altered forms always elicited weaker neuronal responses than the learned forms. This effect was not due to nonsensory factors such as attention, because another sham manipulation, in which one FD was rotated or reflected, did not produce a comparable effect. These results demonstrate that AIT neurones acquire a preference for complex forms through learning by tuning mechanisms.