This study compared cognitive behavioral therapy (CBT) and treatment-as-usual (TAU) in terms of effects on observed social communication-related autism symptom severity during unstructured play time at school for children with autism spectrum disorders (ASD). Thirteen children with ASD (7–11 years old) were randomly assigned to 32 sessions of CBT or community-based psychosocial treatment (TAU) for 16 weeks. The CBT program is based on the memory retrieval competition model and emphasizes the development of perspective-taking through guided behavioral experimentation supplemented with reflective Socratic discussion and supported by parent training and school consultation to promote generalization of social communication and emotion regulation skills. Trained observers blind to treatment condition observed each child during recess on two separate days at baseline and again at posttreatment, using a structured behavioral observation system that generates frequency scores for observed social communication-related autism symptoms. CBT outperformed TAU at posttreatment on the frequency of self-isolation, the proportion of time spent with peers, the frequency of positive or appropriate interaction with peers, and the frequency of positive or appropriate peer responses to the target child (d effect size range 1.34–1.62). On average, children in CBT were engaged in positive or appropriate social interaction with peers in 68.6 % of observed intervals at posttreatment, compared to 25 % of intervals for children in TAU. Further investigation of this intervention modality with larger samples and follow-up assessments is warranted.