Background: To investigate the interpersonal and physical environment mediators of the Transform-Us! mid-intervention effects on physical activity (PA) during recess and lunchtime. Methods: Transform-Us! is a clustered randomised school-based intervention with four groups: sedentary behaviour intervention (SB-I), PA intervention (PA-I), combined PA+SB-I and control group. All children in grade 3 from 20 participating primary schools in Melbourne, Australia were eligible to complete annual evaluation assessments. The outcomes were the proportion of time spent in moderate-to-vigorous PA (MVPA) and light PA (LPA) during recess and lunchtime assessed by accelerometers. Potential mediators included: perceived social support from teachers; perceived availability of line markings; perceived accessibility of sports equipment; and perceived school play environment. Generalised linear models were used and mediation effects were estimated by product-of-coefficients (a·b) approach. Results: 268 children (8.2 years, 57% girls at baseline) provided complete data at both time points. A significant intervention effect on MVPA during recess in the SB-I and PA-I groups compared with the control group (proportional difference in MVPA time; 38% (95% CI 21% to 57%) and 40% (95% CI 20% to 62%), respectively) was found. The perceived school play environment was significantly positively associated with MVPA at recess among girls. An increase in perceived social support from teachers suppressed the PA+SB-I effect on light PA during recess (a·b= −0.03, 95% CI −0.06 to −0.00). No significant mediating effects on PA during recess and lunchtime were observed. Conclusions: A positive perception of the school play environment was associated with higher MVPA during recess among girls. Future studies should conduct mediation analyses to explore underlying mechanisms of PA interventions.