Background: Backward walking (BW) training is increasingly used in rehabilitation for stroke, but relevant evidence remains unclear. Objective: To determine the effect of BW training on patients with stroke. Methods: A keyword search was conducted in PubMed, EMBASE, CINAHL, and China National Knowledge Infrastructure database for articles published until November, 2019. Two investigators screened the articles and extract data from each included study. Meta-analysis was performed to estimate the effect of BW on stroke. In addition, the quality of evidence was evaluated by GRADE (grading of recommendations, assessment, development, and evaluation; version:3.6) approach. Results: A total of ten studies were included according to the inclusion and exclusion criteria in the review. All included studies described some positive influences of BW on stroke relative to the control group (forward walking or conventional treatment). Compared to control group, there is a statistically significant improvement for BW group in gait velocity (mean difference [MD] = 6.87, 95%CI: [1.40, 12.33], P = .01, I2 = 3%), Berg balance score (MD =3.82, 95%CI: [2.12, 5.52], P < .0001, I2 = 0%), and walk test (MD =0.11, 95%CI: [0.02, 0.20], P = .02, I2 = 36%). Conclusions: For patients with stroke, BW training, as an adjunct an adjunct to conventional treatment, can improve Berg balance score (moderate evidence), walk test performance (very low evidence), gait velocity (very low evidence). More large-scale and high-quality studies are warranted.