This study conducted a multilevel meta-analysis to examine the overall effect size of educational robotics in STEM education and the influence of moderator variables. The analysis included 21 studies with 30 effect sizes, covering a total of 2433 participants. The results showed that educational robotics had a moderate-sized positive effect on students' STEM learning, with a mean effect size of 0.488 (95% CI: 0.094–0.882). Specifically, educational robotics had moderate effects on students' learning performances (g=0.665) and learning attitudes (g=0.497), but no significant effect on computational thinking (CT) (g=0.079). Among the moderator variables, discipline was significantly associated with the effects of educational robotics, with technology and cross-disciplinary disciplines showing larger effects compared to science and mathematics. Other moderator variables, such as educational level, instructor support, instructional strategy, interactive type, intervention duration, robotic type, and control group condition, did not significantly influence the effects of educational robotics. The findings provide educational and technological implications for future research and practice in robot-assisted STEM education.This study conducted a multilevel meta-analysis to examine the overall effect size of educational robotics in STEM education and the influence of moderator variables. The analysis included 21 studies with 30 effect sizes, covering a total of 2433 participants. The results showed that educational robotics had a moderate-sized positive effect on students' STEM learning, with a mean effect size of 0.488 (95% CI: 0.094–0.882). Specifically, educational robotics had moderate effects on students' learning performances (g=0.665) and learning attitudes (g=0.497), but no significant effect on computational thinking (CT) (g=0.079). Among the moderator variables, discipline was significantly associated with the effects of educational robotics, with technology and cross-disciplinary disciplines showing larger effects compared to science and mathematics. Other moderator variables, such as educational level, instructor support, instructional strategy, interactive type, intervention duration, robotic type, and control group condition, did not significantly influence the effects of educational robotics. The findings provide educational and technological implications for future research and practice in robot-assisted STEM education.