Studies have reported that physical exercise reduces maladaptive stereotyped motor behaviours (SMB) in children with ASD, but these intervention studies vary in design and outcome. The present systematic review and meta-analysis included 22 studies, involving 274 children with ASD, to quantify the effect of exercise on SMB and its potential moderators. Multi-level modelling revealed a large overall effect, Hedges' "g" = 1.16, with significant heterogeneity across participant, treatment, and study levels. Further, a more appropriate model using between-case estimation for within-subject effects to improve comparability between single-case and group-design studies, yielded a smaller but still significant effect, "g" = 0.51. Lastly, higher exercise intensity, but not age, exercise duration or settings, reliably predicted treatment effectiveness. Implications for clinical practice and research are discussed.