The objective was to examine the effects of backward walking training for improving walking speed and balance in children with cerebral palsy. A systematic review of randomized trials was conducted. Trials had to include children with cerebral palsy, with a Gross Motor Function Classification System, between I and III, that delivered backward walking training as a solo intervention or in combination with forward walking training. The outcomes of interest were walking speed and balance. The methodological quality of included trials was assessed by the PEDro scale, and the quality of evidence was assessed according to Grading of Recommendations Assessment, Development and Evaluation. Eight papers, involving 156 participants, were included. Using random-effects meta-analysis, we estimated that backward walking training improved walking speed by 0.10 m/s [95% confidence interval (CI) 0.05–0.16] and by 2 points on the Pediatric Balance Scale (0–56) (95% CI 1.5–2.2) more than forward walking training. We also estimated that the addition of backward walking training increased walking speed by 0.20 m/s (95% CI 0.07–0.34) and reduced the angular excursion of the center of gravity by 0.5 degrees (95% CI −0.7 to −0.3). The quality of the evidence was classified as low to moderate. In conclusion, overall, backward walking training appears to be as effective or slightly superior to forward walking training for improving walking speed in children with CP. The addition of backward walking training statistically significantly and clinically important enhanced benefits on walking speed.