This meta-analysis included 27 empirical studies to explore the effectiveness of robot-assisted language learning (RALL) as well as the moderating effects of a number of variables. The comparison of 64 effect sizes from 2,637 participants indicated significant variability between the independent samples. The overall results showed that the combined linguistic and affective gain of RALL was significant. The results of moderator analysis showed that most moderators, including place of the study, type of robot, publication type, instructional level of the learners, design of the study, time points of assessment, learning setting, second language (L2) proficiency level, type of assessment, language domain, and the non-RALL condition had no significant effects on learning outcomes. However, the moderating effect of the participants' age was significant as RALL was found to be more effective for elementary school students (7-12 years old). The findings provide insights into the design of effective robots for second language acquisition (SLA) that support learners with certain characteristics.