The purpose of this meta-analysis was to examine the overall effect and impact of moderators in the context of STEM education. This study first summarized previous meta-analysis studies on STEM education and pinpointed gaps in the current literature, such as inconsistent results and the mixed effects of moderators. It also assessed critical features that should be considered in STEM education. The key inclusion criteria were studies on STEM education that used experimental or quasi-experimental research designs and learning outcome measures to evaluate the effect of STEM education. As a result, 23 eligible journal articles published between 2000 and 2020 representing 9,797 participants from 70 schools in seven countries were meta-analyzed. The results demonstrated that the overall mean weighted effect size was 0.38, indicating that STEM education had a small to medium positive effect on students' academic achievements compared with traditional education. In addition, the mean effect size was affected by education levels, subject domains, school locations, learning outcomes, and hours of instruction, but not by pedagogical approaches. Taken together, the results highlight the theoretical framework that will be of value to future STEM education studies as well as the affordances of various moderators for STEM learning.