OBJECTIVE: To summarize the effect of dual-task training on cognitive, physical function, and dual-task performance in people with mild cognitive impairment or dementia. DATA SOURCES: Embase, PEDro, PsycINFO, PubMed, CINAHL, The Cochrane Library, and a forward search conducted via Web of Science have been searched from inception to July 2023. REVIEW METHODS: Good-quality randomized controlled trials compared dual-task training with no/placebo intervention or single-task training among people with a primary diagnosis of mild cognitive impairment or dementia were included. The PEDro scale was used to evaluate the methodological quality of individual studies. The Grading of Recommendations, Assessment, Development and Evaluations system was adopted to appraise the quality of evidence for each outcome. RESULTS: Eighteen trials (1325 participants) were included, and 17 provided data for meta-analysis. Comparing with no intervention, dual-task training led to significant improvements on attention (mean difference (MD) = -20.66, 95%CI [-39.42, -1.90]), functional mobility (MD = -2.73; 95%CI [-3.98, -1.49]). Compared with single-task training, dual-task training had greater effects on overall cognitive function (standardized mean difference (SMD) = 0.29, 95%CI [0.09, 0.49]), balance (SMD = 0.78, 95%CI [0.40, 1.15]) and functional mobility (MD = -1.17; 95%CI [-1.77, -0.58]). Its effect on dual-task performance remains inconclusive due to the inconsistent results reported. CONCLUSION: Low- to moderate-quality evidence supports that dual-task training has beneficial effects on cognitive function and physical function in individuals with dementia or mild cognitive impairment. The optimal training protocol of dual-task training on cognitive and physical functions, and dual-task performance remains uncertain. Well-designed, randomized studies with large enough sample sizes are warranted.