Theoretical models of reading comprehension have consistently highlighted the importance of inference generation in the meaning-making process. Additionally, previous research has indicated that instruction in making inferences is effective at improving inference ability and general reading comprehension. In this meta-analysis, we aimed to further examine the effectiveness of inference instruction considering the characteristics of participants, instruction, and outcomes that may have an impact on the effects. We identified 56 experimental and quasi-experimental studies (N = 5,088), including 81 independent samples and 138 effect sizes. Using robust variance estimation, inference instruction had a moderate, positive effect (g = 0.495, p < 0.001) on student inference skills and general comprehension. No significant differences in the effectiveness of the inference instruction were found across participant characteristics (grade, reading ability, and multilingual status) or outcome measures (type of measure or number of inferences assessed). For instructional characteristics, no significant differences were identified based on group size, the type of practice, the number of inferences taught, or the type of text. However, there was a statistically significant difference suggesting that reading text aloud to students (g = 0.28) may be less effective than when students read the text independently (g = 0.79) during instruction. Finally, study quality was evaluated as a moderator using six characteristics of quality, and no significant differences in effect size were found between studies of higher quality and those of lower quality. Thus, this review suggests that regardless of age, reading ability, type of instruction, and text, inference instruction would be worthy of implementation.