Flipped learning has become a popular approach in various educational fields, including second language teaching. In this approach, the conventional educational process is reversed so that learners do their homework and prepare the material before going to class. Class time is then devoted to practice, discussion, and higher-order thinking tasks in order to consolidate learning. In this article, we meta-analysed 56 language learning reports involving 61 unique samples and 4,220 participants. Our results showed that flipped classrooms outperformed traditional classrooms, g = 0.99, 95% CI (0.81, 1.17), z = 10.90, p < 0.001. However, this effect had high heterogeneity (about 86%), while applying the Trim and Fill method for publication bias made it shrink to g = 0.58, 95% CI (0.37, 0.78). Moderator analysis also showed that reports published in non-SSCI-indexed journals tended to find larger effects compared to indexed ones, conference proceedings, and university theses. The effect of flipped learning did not seem to vary by age, but it did vary by proficiency level in that the higher proficiency the higher the effects. Flipped learning also had a clear and substantial effect on most language outcomes. In contrast, whether the intervention used videos and whether the platform was interactive did not turn out to be significant moderators. Meta-regression showed that longer interventions resulted in only a slight reduction in the effectiveness of this approach. We discuss the implications of these findings and recommend that future research moves beyond asking whether flipped learning is effective to when and how its effectiveness is maximized.