Although a number of studies have demonstrated that psychiatric patients' performance deficits on the Wisconsin Card Sorting Test (WCST) can be modified through intervention, relatively little evidence exists to support the long-term durability of training effects. The present study tested the effectiveness and durability of a training procedure based on errorless learning principles, and in addition sought to determine the effect of previously committed errors on training and post-training performance. Twenty-three chronic psychotic inpatients were randomly assigned to an Initial Error (n = 11) or No Initial Error (n = 12) group. The Initial Error group received two standard administrations of the WCST prior to training (where they were expected to commit many errors); the No Initial Error group had no prior exposure to the WCST. All subjects received training on the WCST which was followed by immediate, 1-, 2-, and 4-week post-tests. Results supported the effectiveness of training and the durability of effects, but previous error history showed no clear relationship to post-training performance.