This meta-analysis examines the effects of game-based learning (GBL), compared with traditional instructional methods, on the science achievement of students, ranging from primary school to university. Results from the 41 primary studies of 6256 participants showed heterogeneity among effect sizes. The random effects model results showed a positive effect (g = 0.705, 95% CI = [0.603, 0.807], p < 0.001), suggesting that students learned substantially more via GBL than traditional instruction. Additionally, moderation analysis found that the link between GBL and science achievement was stronger (a) among students in Eastern countries than Western countries; (b) among primary school students than undergraduates, junior or senior secondary school students; (c) on quizzes, final exams, and course grades than assignment scores; (d) for interventions between 4 hours and 1 week, smaller for those less than 4 hours, and smallest for those over 1 week; and (e) slightly in later publication years.