Mental simulation of future scenarios is hypothesized to affect future behavior, but a large and inconsistent literature means it is unclear whether, and under what conditions, mental simulation can change people’s behavior. A meta-analysis was conducted to synthesize the effects of mental simulation on behavior and examine under what conditions mental simulation works best. An inclusive systematic database search identified 123 (N = 5685) experiments comparing mental simulation to a control group. After applying a multilevel random effects model, a statistically-reliable positive effect of Hedges’ g=0.49 [95% CI 0.37; 0.62], which was significantly different than zero. Using a taxonomy to identify different subtypes of mental simulation (along two dimensions, class [process, performance, outcome] and purpose [whether an inferior, standard, superior version of that behavior is simulated]), it was found that superior simulations garnered more reliable beneficial effects than inferior simulations. These findings have implications for integrating theories of how mental simulations change behavior, how mental simulations are classified, and may help guide professionals seeking evidence-based and cost-effective methods of changing behavior.