Despite their social withdrawal, school-age children with autism spectrum disorder who are minimally verbal (MVASD; i.e., use a limited repertoire of communicative spoken words) have received few interventions supporting peer engagement. This study examined efficacy of a novel ecological randomized controlled trial—school-based peer social intervention—designed to increase social engagement (via available communication channels) in school-age peer dyads with MVASD. Fifty-four children with MVASD (8–16 years) in 9 Israeli special education schools were randomly assigned to conversation intervention, collaboration intervention, or waitlisted treatment-as-usual (control) group (n = 18 per group). Manualized conversation and collaboration interventions each included 60 lessons (15 weeks × 4), implemented by teachers at school and supervised by researchers. Pretest–posttest improvement in spontaneous peer interaction was measured via 3 data sources/methods: teacher-reported social behavior (Vineland: Socialization domain) and direct observations of children’s spontaneous free play (Modified-Classroom Observation Schedule to Measure Intentional Communication) and free conversation (Social Conversation Scale). Allocation group was masked from reporters/coders. As secondary outcomes, children’s progress was measured in executive functions (BRIEF Inventory), and communication (Vineland). Significant pre–post improvement emerged for both intervention groups’ spontaneous free conversation and for the collaboration group’s spontaneous free play. Teacher reports, although mixed, indicated that the conversation group’s socialization skills improved, but communication did not. Children in the conversation group also improved their metacognitive executive skills (e.g., planning, monitoring, organization). Strengthening this high-risk school-age population’s ability to interact more spontaneously with peers through conversation and collaboration intervention holds promise for reducing social withdrawal in MVASD.