Teachers are increasingly using augmented reality (AR) to develop students' higher-order thinking (HOT). As past studies showed mixed results, our "random effects meta-analysis" of 21 effect sizes from 17 studies of 1256 participants determined the overall effect of AR on HOT and accounted for differences across studies via moderator effects. AR had a large positive overall effect on HOT (g = 1.109, 95% CI [0.699, 1.518]) but differed significantly across studies. AR effects on HOT were (a) largest for 3D visualization (g = 1.216), with progressively smaller ones for immersion (g = 1.198), collaboration (g = 0.934) and contextualization (g = 0.408), (b) larger for skills (g = 1.243) than disposition (g = 0.371), and (c) larger for convergent HOT (g = 1.625) than divergent HOT (g = 0.395).