Proponents of the verbal behavior approach to instruction for individuals with autism have identified mand training as a starting point for early intervention. Mand training is a process whereby the learner is taught to request highly preferred items under conditions when those items are most valuable. A hypothesized benefit of this approach is that mand training has a collateral effect on nontargeted behavior, though empirical support for this hypothesis is currently tenuous. The present investigation examined the collateral effects of vocal mand training compared to vocal request training for 3 previously nonvocal children with autism. Levels of orienting toward a speaker, compliance with instruction, and overall problem behavior were measured across experimental conditions and analyzed using an alternating treatment design. Results indicate that problem behavior for all participants occurred at lower levels during mand training than request training and that the level of nontargeted social behavior (i.e., orienting and compliance) was higher during mand training for 2 participants and was similar across both conditions for the final participant. The results support the hypothesis that mand training can have a collateral impact on nontargeted behavior, though explicit instruction of social behaviors may also need to be embedded within mand training procedures. (Contains 2 tables and 3 figures.)