Game‐based learning (GBL) may address the unique characteristics of a single subject such as chemistry. Previous systematic reviews on the effects of GBL have yielded contradictory results concerning cognitive and motivational outcomes. This meta‐analysis aims to: (a) estimate the overall effect size of GBL in chemistry education on cognitive, motivational, and emotional outcomes compared with non‐GBL (i.e., media comparison); (b) examine possible moderators of the effects; and (c) identify the more effective game design and instructional design features (i.e., value‐added comparison). We screened 842 articles and included 34 studies. This study is the first GBL meta‐analysis that employed a three‐level random‐effects model for the overall effects. Moderator analysis used a mixed‐effects meta‐regression model. Results from the media comparison suggest chemistry GBL was more effective for cognition (g = 0.70, k = 30, N = 4155), retention (g = 0.59, k = 20, N = 2860), and motivation (g = 0.35, k = 7, N = 974) than non‐GBL and the substantial heterogeneity (I2 = 86%) for cognitive outcomes. No study reported emotional outcomes, and studies considering value‐added comparisons of GBL with versus without specific design features (k = 3) were too few to perform a meta‐analysis. Moderator analyses implied that except for publication source and sample size, no other moderator was related to effect sizes. There may be the small‐study effects, particularly publication bias. Although we conclude that GBL enhances chemistry learning more than non‐GBL, the results also make clear that additional high‐quality value‐added research is needed to identify design guidelines that may further improve chemistry GBL. More GBL meta‐analyses on subjects other than chemistry are also needed. As the first GBL meta‐analysis that emphasizes emotion, we call for more research on emotion and on relationships between cognition, motivation, and emotion in GBL.