The accurate measurement of neurocognitive function requires stable and reliable instruments. These assessments are critically important as correlates, predictors, and outcome measures of psychopathology, neuropathology, and treatment. Particularly, in studies that evaluate the impact of treatments on neurocognitive function, it is essential that their reliability as repeated measures be quantified. In an attempt to measure within- and between-subject variance on a battery of neuropsychological tests commonly used in schizophrenia research, baseline measures were administered twice, approximately 10 weeks apart, to 54 stable outpatients with schizophrenia or schizoaffective disorder. Instruments were judged for their stability, and individuals were assessed for individual differences in test-retest variability. The majority of the instruments administered were highly stable. Group means for several tests showed improvement on retest, which may indicate a practice effect, although many tests did not reach our criteria for "change." Of 962 test scores, 21 (2%) showed significant change. Most subjects were stable, with only 7% of subjects accounting for 38% of significant test-retest changes. Results indicate that the instruments assessed have sufficient reliability and stability to be used as repeated measures in clinical trials and that a small number of patients may have instability of performance as an individual characteristic.