Background : It was suggested that robot-assisted gait training (RAGT) should not be routinely provided to disabled patients in place of conventional over-ground walking training (CGT). There exist several randomised controlled trials reporting on RAGT for people with multiple sclerosis. However, the effectiveness of RAGT varies between studies with the effectiveness pointing in different directions. It might be possible that the effectiveness of RAGT and CGT depends on the disease related disabilities of the people included in the clinical studies. We aimed to systematically search RCTs and to perform a meta-regression to compare the effects of robot-assisted gait training in people with less and higher disease related disabilities. The Expanded Disability Status Scale (EDSS) scores were used to classify level of disability. Methods : A systematic search was developed to search four electronic databases (MEDLINE, CENTRAL, EMBASE and CINAHL) for eligible articles. A random effects model was applied to meta-analyse the effects of the interventions. Meta-regression was performed with an uni-variable random effects model using baseline walking speed and EDSS to predict the between group effect. Results : The search on databases resulted in 596 records and finally nine studies were included into the review. The pooled estimates of the effects for performance over short and long distance tests were small and non-significant: -0.08 SMD (95% CI: -0.51 to 0.35) and − 0.24 SMD (95% CI: -0.67 to 0.19). Neither baseline walking speed or disease related disability were related to the mean effect size. Discussion : Future studies are needed to help clinicians to decide, which intervention should be allocated to the individual patient.