A meta-analysis of randomized controlled trials (RCTs) has recently showed that theory-based interventions designed to promote physical activity (PA) significantly increased PA behavior. The objective of the present study was to investigate the moderators of the efficacy of these theory-based interventions. Seventy-seven RCTs evaluating theory-based interventions were systematically identified. Sample, intervention, methodology, and theory implementation characteristics were extracted, coded by three duos of independent investigators, and tested as moderators of interventions effect in a multiple-meta-regression model. Three moderators were negatively associated with the efficacy of theory-based interventions on PA behavior: intervention length (=14 weeks; ß = -0.22, p = 0.004), number of experimental patients (ß = -0.10, p = 0.002), and global methodological quality score (ß = -0.08, p = 0.04). Our findings suggest that the efficacy of theory-based interventions to promote PA could be overestimated consequently due to methodological weaknesses of RCTs and that interventions shorter than 14 weeks could maximize the increase of PA behavior.