The primary goal of this study was to investigate transfer of training (generalization) in patients with schizophrenia. We randomly assigned 33 schizophrenia subjects to one of three conditions: training on the Wisconsin Card Sort Test (WCST-T), training on the Halstead Category Test (CAT-T), or no training (No-T). The WCST and CAT were administered to all subjects at baseline. Subjects in the WCST-T and CAT-T groups then received training on the respective test, while the No-T group received additional untrained trials. All participants were subsequently retested on the WCST and CAT, and completed a brief neuropsychological battery. As hypothesized, the WCST-T and CAT-T groups exhibited large improvements on the trained test and moderate improvement on the untrained test, while the No-T group failed to show improvement on either test. These results suggest that the training paradigm did produce generalization, and that the changes were not due to practice effects. The extent of generalization across both training groups was strongly associated with neuropsychological test performance (Spearman's rho=0.56, P<0.05). The implications of these findings for rehabilitation programs were discussed, and recommendations were made for future research.