Studies of procedural learning in medicated schizophrenia patients using predictive saccade paradigms have consistently demonstrated hypometric predictive responses. Findings from antipsychotic-naive schizophrenia patients indicate fewer or no deficits. This pattern of findings suggests that antipsychotic medications might adversely affect frontostriatal systems supporting procedural learning on this task. The accuracy and latency of predictive saccades were assessed in 25 antipsychotic-naive first-episode schizophrenia patients and 22 matched healthy individuals. Patients were retested after 6 weeks of treatment with risperidone. Healthy individuals were reevaluated after a similar time period. The ability to learn to time response initiation in anticipation of target appearance (target prediction) was not impaired in patients before or after treatment. In contrast, although no deficits were evident before treatment initiation, after treatment patients showed a marked decrease in the accuracy of predictive but not sensory-guided responses. The findings from pretreatment testing indicate that procedural learning is a relatively unaffected cognitive domain in antipsychotic-naive first-episode schizophrenia. Although treatment-emergent extrapyramidal symptoms were minimal, these data suggest that D2 antagonism in striatum after risperidone treatment was sufficiently robust to disrupt the generation of planned volitional behavior guided by internalized representations.