Background and Objective: In Parkinson's disease (PD), wearing off and side effects of long-term medication and complications pose challenges for neurologists. Although Tai Chi is beneficial for many illnesses, its efficacy for PD remains uncertain. The purpose of this review was to evaluate the efficacy and safety of Tai Chi for PD. Methods: Randomized controlled trials (RCTs) of Tai Chi for PD were electronically searched by the end of December 2013 and identified by two independent reviewers. The tool from the Cochrane Handbook 5.1 was used to assess the risk of bias. A standard meta-analysis was performed using RevMan 5.2 software. Results: Ten trials with PD of mild-to-moderate severity were included in the review, and nine trials (n = 409) were included in the meta-analysis. The risk of bias was generally high in the blinding of participants and personnel. Improvements in the Unified Parkinson's Disease Rating Scale Part III (mean difference (MD) -4.34, 95% confidence interval (CI) -6.67 - -2.01), Berg Balance Scale (MD: 4.25, 95% CI: 2.83-5.66), functional reach test (MD: 3.89, 95% CI: 1.73-6.04), Timed Up and Go test (MD: -0.75, 95% CI: -1.30 - -0.21), stride length (standardized MD: 0.56, 95% CI: 0.03-1.09), health-related quality of life (standardized MD: -1.10, 95% CI: -1.81- -0.39) and reduction of falls were greater after interventions with Tai Chi plus medication. Satisfaction and safety were high. Intervention with Tai Chi alone was more effective for only a few balance and mobility outcomes. Conclusions: Tai Chi performed with medication resulted in promising gains in mobility and balance, and it was safe and popular among PD patients at an early stage of the disease. This provides a new evidence for PD management. More RCTs with larger sample size that carefully address blinding and prudently select outcomes are needed.