<p>This study compares the effects of metaverse and face-to-face (F2F) classroom settings on the speaking performance of Korean middle school English as a Foreign Language (EFL) learners. Specifically, it examines how affective enjoyment and anxiety influence speaking outcomes through participation and social interaction in each context. Two hundred seventy-two students were assigned to either a F2F or metaverse environment (136 students per group). Mixed design ANOVA revealed that, although speaking proficiency improved in both settings, the metaverse group showed higher posttest scores, reported more enjoyment, and had more social interactions. The F2F group, however, had descriptively higher participation, although the difference was not statistically significant. A moderated mediation analysis revealed different learning pathways for each environment. In the F2F setting, participation was the strongest predictor of learning outcomes, with enjoyment indirectly supporting gains primarily through participation. Conversely, in the metaverse, social interaction was the main driver of performance, especially for learners with medium-to-high anxiety, and mediated the impact of enjoyment. These findings suggest that the two settings facilitate second language speaking through different mechanisms of engagement, although direct causal attributions require caution. The study highlights the potential for instructional approaches tailored to the unique features of each setting when teaching second-language speaking to young beginner-level EFL students.</p>