Objective: To examine the effectiveness of rehabilitation training based on virtual reality in improving balance, quality of life, activities of daily living, and depressive symptoms of patients with Parkinson’s disease. Data sources: PubMed, EMBASE, CINAHL, Scopus, Cochrane Library, PsycINFO, ProQuest, Physiotherapy Evidence Database, IEEE Xplore, China National Knowledge Infrastructure, Wanfang, and VIP Information databases were searched from their inception to October 15, 2020. Trial registries, gray literature, and target journals were also searched. Methods: Eligible randomized controlled trials included studies with patients with Parkinson’s disease in rehabilitation training based on virtual reality. Comprehensive Meta-Analysis 3.0 software was used. Physiotherapy Evidence Database Scale and the Grading of Recommendation, Assessment, Development, and Evaluation system were used to assess the methodological quality of individual trials and the overall quality of the evidence, respectively. Results: A total of 22 randomized controlled trials with 836 patients were included. Meta-analysis revealed that training significantly improved balance ( g = 0.66, P < 0.001), quality of life ( g = 0.28, P = 0.015), activities of daily living ( g = 0.62, P < 0.001), and depressive symptoms ( g = 0.67, P = 0.021) compared to the control group. Subgroup analysis indicated that training should utilize video game consoles. Meta-regression analyses showed that age, sessions, and frequency of training had statistically significant impacts on balance scores. Quality of individual trials was high and overall evidence ranged from very low to low. Conclusion: Virtual rehabilitation training could be adopted in healthcare institutions as supplementary training for patients with Parkinson’s disease.