Play is, by definition, internally motivated, flexible, spontaneous and voluntary. Yet some researchers claim to have taught children with autism to play using behavioural interventions that are heavily structured, repetitive and make use of external reinforcements. In the current systematic review, we examine the extent to which these claims are supported by the evidence presented by the researchers themselves. We conclude that the most effective behavioural interventions have been those which have built on children's existing abilities or have relied on the motivating nature of activities themselves rather than on external rewards. We discuss the problems inherent in distinguishing between behavioural and cognitive change in children's play and highlight generalization as a poorly understood but focal process. Finally, we discuss the value of teaching children with autism play behaviours when these are not characterized by the defining qualities of play as a disposition.