Conditional rates of problem behavior for weeks that followed medication decreases and no medication changes were compared for 12 individuals who exhibited severe problem behavior (e.g., self-injury, aggression). The results indicate that conditional rates of problem behavior were higher following medication decreases than following no changes in medication. During the subsequent week, rates of problem behavior typically decreased without reinstating the prior dosage of medication. Additional analyses suggest that the first medication decrease typically produced the highest increase in problem behavior. Decreases in medication also disrupted sleep patterns for several individuals. Possible operant conceptualizations of behavior changes produced by medication decreases are discussed. (Contains 1 table and 4 figures.)