Social learning through friendships is important in child development. Autistic children often initiate and engage in social interactions differently than neurotypical peers. LEGO® based therapy is a group intervention which facilitates social interactions with peers using collaborative LEGO® play. A 1:1 cluster randomised controlled trial with autistic children aged 7-15 comparing 12 weeks' LEGO® based therapy and usual support to usual support alone in 98 mainstream schools (2017-2019) was carried out. The primary outcome was social skills (Social Skills Improvement System) completed by unblinded teachers (not intervention deliverers) at 20 weeks. Analysis used intention-to-treat. Fifty intervention schools (127 participants) and 48 control schools (123 participants) were allocated. Primary analysis included 45 intervention schools (116 participants) and 42 control schools (101 participants). The between-arms difference was 3.74 (p = 0.06, 95% confidence interval: -0.16, 7.63) and 1.68 (p = 0.43, 95% confidence interval: -2.51, 5.87) at 20 and 52 weeks (0.18 and 0.08 standardised effect sizes). Twenty-week outcomes for those receiving per protocol intervention were 4.23 (95% confidence interval 0.27, 8.19) with a standardised effect size of 0.21. Sensitivity estimates were between 3.10 and 4.37 (0.15-0.21 standardised effect sizes). Three unrelated serious adverse events were reported. LEGO® based therapy has a small positive non-significant benefit for social skills at 20 weeks but not 52 weeks.