Two groups of schizophrenic patients took the Wisconsin Card Sorting Test (WCST) in separate follow-up studies; a naturalistic clinical follow-up (n = 12, study 1) and a neuroleptic reduction study (n = 10, study 2). 11 of 22 subjects (50%) performed well on the task at one or both time points. In each study no group differences were found between time 1 and time 2 performance on three WCST variables. However, correlational analyses revealed considerable within-subject variation on the WCST in the study 1 sample, which was composed of younger and more acute patients than those in study 2. This variation was present despite within-subject stability on the WAIS-R Vocabulary and Block Design subtests. For the sample combined, a trend was found (p = 0.12) linking better WCST performance at either time period with higher WAIS-R Vocabulary scores. This intra-subject variability may reflect fluctuations in neuropsychological performance in schizophrenics who maintain the residual capacity to do the task. These findings highlight the importance of longitudinal studies of neuropsychological functioning in schizophrenia. Studies of larger samples are needed to confirm these initial results.