To investigate the effectiveness of robot-assisted therapy on balance function in stroke survivors. PubMed, the Cochrane Library, Embase and China National Knowledge Infrastructure databases were searched systematically for relevant studies. Randomized controlled trials reporting robot-assisted therapy on balance function in patients after stroke were included. Information on study characteristics, demographics, interventions strategies and outcome measures were extracted by 2 reviewers. A total of 19 randomized trials fulfilled the inclusion criteria and 13 out of 19 were included in the meta-analysis. Analysis revealed that robot-assisted therapy significantly improved balance function assessed by berg balance scale (weighted mean difference (WMD) 3.58, 95% confidence interval (95% CI) 1.89-5.28, p < 0.001) compared with conventional therapy. Secondary analysis indicated that there was a significant difference in balance recovery between the conventional therapy and robot-assisted therapy groups in the acute/subacute stages of stroke (WMD 5.40, 95% CI 3.94-6.86, p < 0.001), while it was not significant in the chronic stages. With exoskeleton devices, the balance recovery in robot-assisted therapy groups was significantly better than in the conventional therapy groups (WMD 3.73, 95% CI 1.83-5.63, p < 0.001). Analysis further revealed that a total training time of more than 10 h can significantly improve balance function (WMD 4.53, 95% CI 2.31-6.75, p < 0.001). No publication bias or small study effects were observed according to the Cochrane Collaboration tool. These results suggest that robot-assisted therapy is an effective intervention for improving balance function in stroke survivors.