Video modeling is an evidence-based instructional strategy for teaching a variety of skills to individuals with autism. Despite the effectiveness of this strategy, there is some uncertainty regarding the conditions under which video modeling is likely to be effective. The present investigation examined the differential effectiveness of video modeling for teaching peer-directed initiations to three young children with autism using a single subject reversal design. In one condition, sharing toys, participants were taught to initiate with peers in the form of inviting a peer to join an activity that was preferred for the participant. In the second condition, joining a play in progress, participants were taught to ask if they could join in a preferred activity with peers. Participants demonstrated initiations during the joining but not during the sharing conditions. The results suggest video modeling can be differentially effective as a function of the target behavior and corresponding environmental events that are both observed and experienced by the child. These outcomes may have implications for how practitioners select social targets and develop video modeling materials when teaching individuals with autism.