Previous empirical studies on the effect of scaffolding in game-based learning environments have shown inconsistent findings. In this study, we conducted a meta-analysis to examine the effect of scaffolding in digital game-based learning (DGBL) and to explore a range of moderating factors that may have contributed to the inconsistencies of primary studies. We used the three-level meta-analysis method to analyze the data for handling data non-dependency issues of multiple effect sizes in one study. A total of 49 primary studies and 154 effect sizes were identified through systematic literature search. The results show that scaffolding in DGBL could effectively improve learning (g = 0.43, 95% CI: [0.30, 0.56]), and the heterogeneity among studies was significant (Q = 708.99, p < 0.001). Furthermore, our results indicate that the studies involving elementary school students and university students showed larger effect sizes of scaffolding than those involving secondary school students. Scaffolding also appeared to be differentially effective in different types of games: more effective in adventure, puzzle, and simulation games than in role-playing and strategy games. Future studies should pay attention to the design of scaffolding mechanisms in educational digital games and to the influence of scaffolding on behavioral patterns and learning processes of learners engaged in DGBL.