In recent years, an increasing number of teachers and researchers have used virtual reality (VR) to enhance English as a foreign language (EFL) learning, but the learning effects they found varied. Because of these differences, we conducted a meta-analysis. This study aims to compare the effectiveness of VR-based EFL learning methods with traditional EFL learning methods, and to determine what factors led to these results. We searched for articles published from 2015 to 2021 using Web of Science, ERIC, and Google Scholar. A total of 23 randomized controlled trials (RCTs) or quasi-experimental studies were included, and the effect size (ES) was calculated. The results show that the VR-based EFL learning method is significantly better than the traditional EFL learning method and has a small positive effect (g = 0.445). We also coded the characteristics of the samples to examine their moderating effect on the results. We found that educational levels, country/area, and EFL learning outcomes have significant differences that can explain the variance in ES among the samples. According to the analysis results, we also provide suggestions on which learning outcomes are most helpful, how to choose the best educational stages, learning materials, and type of VR to promote EFL learning performance, and directions for future research and practice.