Background: Peer relationships improve for children with autism spectrum disorder (ASD) in clinic-based social skills groups but rarely generalize to real world contexts. This study compares child outcomes of two social skills interventions conducted in schools with children in Kindergarten through fifth grade. Method: Children with ASD were randomized to one of two interventions that varied on group composition (mixed typical and ASD vs. all ASD or social difficulties) and intervention approach (didactic SKILLS based vs. activity-based ENGAGE groups). Interventions were implemented at school for 8 weeks (16 sessions) with an 8-week follow-up. Innovative measures of peer nomination and playground peer engagement, as well as teacher reports of child behavior problems and teacher-child relationship were analyzed for 137 children with ASD across four sites. Results: On the primary outcome of social network connections from the peer nomination measure, there was no main effect of treatment, but there were moderator effects. Children with low teacher-child closeness or high conflict improved more in their social connections if they received the SKILLS intervention, whereas children with higher teacher-child closeness improved more if they received the ENGAGE intervention. Only two secondary outcome measures yielded significant effects of treatment. Children in the SKILLS groups increased peer engagement and decreased isolation during recess. Child behavior problems and teacher-child closeness moderated peer engagement such that children with higher behavior problems and lower closeness benefited more from SKILLS groups. Conclusions: These findings suggest that social skills groups conducted at school can affect both peer engagement during recess as well as peer acceptability. Child characteristics and teacher-child relationship prior to intervention yield important information on who might benefit from a specific social skills intervention.