This meta-analysis examined whether learning outcomes differ (a) for STEM integration versus traditional instruction and (b) across STEM integration implementations. Based on 79 effect sizes from 40 studies of 15,577 students, those learning via STEM integration outperformed other students on academic achievement tests (g = 0.661; 95% CI [0.548, 0.774]). The effect sizes of STEM integration on achievement were largest for context integration, smaller for content integration, and smallest for tool integration. They were largest for inquiry-based learning, and progressively smaller for problem-based learning, designed-based learning, and project-based learning. They were largest for STEM subject achievement, and progressively smaller for science achievement, math achievement, and engineering achievement. They were larger for collectivist countries than for individualistic countries. Engineering design skills and grade level were not significant moderators. These results can inform integrated STEM instructional design and improve student learning.