Generative drawing, or drawing-to-learn, has been studied as an effective learning strategy primarily used in science learning. However, drawing can be cognitively demanding, requiring instructional support to ensure good learning outcomes. The present meta-analysis encompassed 14 studies with 16 statistically independent comparisons between unsupported and instructionally supported drawing groups regarding differences in comprehension and transfer as learning outcomes (total N = 1213 participants). Potential moderators of drawing support's effectiveness included "reduction of extraneous cognitive load" (e.g., providing a partially drawn background, thus reducing the cognitive demands of drawing mechanics), "increase of germane cognitive load" (e.g., asking students to write an explanation before drawing), and "facilitation of integration" (e.g., asking students to compare their drawings with provided illustrations). Surprisingly, drawing support in general was not related to students' learning outcomes. However, "facilitation of integration" turned out to be a significant moderator: While drawing support that could not be expected to facilitate the integration process did not significantly affect students' comprehension (g = - 0.15), drawing support that could be expected to facilitate the integration process significantly increased students' comprehension (g = 0.35). The two other moderators were not significant. These results are in line with generative learning theory and highlight the specific relevance of the integration sub-process of generative learning: After selecting and organizing new information in verbal and pictorial models, it seems essential to help students make sense by integrating the verbal and pictorial models into a coherent mental model that can consistently be stored in long-term memory.