Immersive virtual reality (VR) produces moderate, reliable gains in both language proficiency and affective engagement relative to conventional instruction, based on a systematic review and meta-analysis of 21 quantitative studies (N = 1,144) published between 2010 and 2021. Pooled effect sizes, estimated using Hedges' g under a random-effects model, were 0.662 (95% CI [0.398–0.925], p < 0.001) for linguistic outcomes and 0.570 (95% CI [0.309–0.831], p < 0.001) for affective outcomes. Substantial heterogeneity was detected for linguistic outcomes (I² = 75.5%), prompting moderator analyses across five dimensions: educational level, hardware type, target language skill, target language, and L1/L2 context. Fully immersive systems (head-mounted displays, CAVE) yielded larger effects than desktop or mobile platforms, and vocabulary and writing showed stronger gains than listening. No moderator reached statistical significance, indicating that implementation quality and design factors likely account for the observed variance. Publication bias was unlikely, supported by symmetrical funnel plots and large fail-safe N values. Findings are limited by the geographic concentration of studies (predominantly East Asia), the dominance of quasi-experimental designs, and short intervention durations. Immersive VR represents a credible, evidence-based complement to language instruction, though longitudinal trials across diverse contexts are needed before broad implementation can be recommended.