Augmented reality (AR) continues to show its impact in education. While systematic review and meta-analysis studies have synthesized the evidence of this impact, most of them did not differentiate how AR may foster various learning outcomes to different degrees. In this study, 134 (quasi-)experimental studies on augmented reality (AR) in education from 2012 to 2021 were reviewed to discern the impact of AR on three levels of learning outcomes, namely response, knowledge and skill, and performance. The 134 studies are the data of this study for meta -analysis, which were obtained following the procedure of the Preferred Reporting Items for Systematic reviews and Meta-Analysis. Moreover, meta-regression was employed to investigate important factors relating to the variation of the impact, including educational level, subject area, treatment duration, and AR affordance. Findings from the meta-analysis suggest that AR tech-nology has benefited all three learning outcomes, with a larger mean effect size on the outcome of performance. Furthermore, the meta-regression results indicate treatment duration as a signifi-cant factor likely relating to the variation in the impact of AR in education. Moreover, AR used to support the learning of languages or social studies is likely associated with higher positive learner responses such as learning motivation or attitude than AR used for science learning. The meta -regression result indicated that use of 3-D visualization in AR needs to be carefully designed and evaluated. The implications of using AR to support learning, the design of AR learning en-vironments, and the future direction of research are discussed.