Elementary school children with normal intelligence and autism (n = 20), mental retardation and no autism (n = 24), and typical development (n = 37) were observed for 15 minutes during recess at school. Ten behaviors were scored as present or absent using the Playground Observation Checklist. Children with autism were distinguished from children with mental retardation and typical development by their social problems, whereas children with typical development and mental retardation did not differ significantly in social competency. The four social behaviors on the checklist correctly identified 94 percent of the children as having or not having autism. All of the children with autism and all of the typical children were correctly classified. Our pilot findings suggest that the Playground Observation Checklist has potential as a simple and clinically useful component of a comprehensive evaluation for possible autism. (Contains 1 table.)