Objective To assess the effects of dual-task training on gait and balance in stroke patients. Data sources: A systematic review of PubMed, Web of Science, Embase and Cochrane Library from their inception through 20 August 2021. Review methods The bibliography was screened to identify randomized controlled trials that applied dual-task training to rehabilitation function training in stroke patients. Two reviewers independently screened references, selected relevant studies, extracted data and assessed risk of bias using the Cochrane tool of bias. The primary outcome was the gait and balance parameters. Results A total of 1992 studies were identified and 15 randomized controlled trials were finally included (512 individuals) were analyzed. A meta-analysis was performed and a beneficial effect on rehabilitation training was found. Compared to patients who received conventional rehabilitation therapy, those who received dual-task training showed greater improvement in step length (MD = 3.46, 95% CI [1.01, 5.92], P = 0.006), cadence (MD = 4.92, 95% CI [3.10, 6.74], P < 0.001) and berg balance scale score (MD = 3.10, 95% CI [0.11, 6.09], P = 0.040). There were no differences in the improvements in gait speed (MD = 2.89, 95% CI [ − 2.02, 7.80], P = 0.250) and timed up and go test (MD = −2.62, 95% CI [ − 7.94, 2.71], P = 0.340) between dual-task and control groups. Conclusion Dual-task training is an effective training for rehabilitation of stroke patients in step length and cadence, however, the superiority of dual-task training for improving balance function needs further discussion.