Virtual reality (VR) has become a promising tool for enhancing social skills in special education, K-12 education, and post-secondary education. However, there is a lack of comprehensive meta-analyses examining the effectiveness of VR social skills training and the factors that moderate its efficacy. In this study, a meta-analysis approach was used to quantitatively analysis 31 experimental or quasi-experimental studies, systematically examining the effectiveness of VR social skills training. The results indicate that VR training is an effective method for improving social skills, with a statistically significant overall effect size (ES = 0.667). However, the effectiveness of VR training varies depending on the complexity of the social skills being targeted. While VR training significantly enhances trainees' complex social skills (ES = 0.984), it does not have a significant effect on basic social skills (ES = 0.133). Furthermore, the intervention duration and interaction methods were found to be important factors influencing the efficacy of social skills training. These findings suggest that VR social skills training can be an effective approach, particularly for complex social skills, and that intervention duration and interaction method should be carefully considered when designing VR training programs. The paper concludes with a discussion of the results and provides directions for future research.