Background: Cognitive impairment is common, debilitating, and poorly managed in persons with multiple sclerosis (pwMS). Exercise training might have positive effects on cognitive performance in pwMS, yet the overall magnitude, heterogeneity, and potential moderators remain unclear. Objective: This three-level meta-analysis aims to identify the effects of exercise training and those of exercise modalities on global and domain-specific cognitive performance in pwMS. Methods: MEDLINE, PsycInfo, SportDiscus, CENTRAL, and EMBASE were screened for randomized and non-randomized clinical trials from inception to 27 January 2020, yielding 3091 articles. Based on titles and abstracts, 75 articles remained in the selection process. After full-text evaluation, 13 studies were finally selected (PROSPERO pre-registered). Results: The pooled effect of exercise training on the global cognitive performance was null ( g = 0.04, 95% confidence interval (CI): –0.11 to 0.18) and no significant differences were displayed among domains. Heterogeneity within studies was null ([Formula: see text]= 0.0%) and between studies was low ([Formula: see text]= 25.1%). None of the moderators (exercise modalities, age, Expanded Disability Status Scale (EDSS), supervision, cognitive domains) reached significance. However, the exercise volume explained most of the overall heterogeneity (slope = 4.651 × 10⁻⁵, [Formula: see text] = 100%, [Formula: see text] = 52.34%). Conclusion: These results do not support the efficacy of exercise training on global or domain-specific cognitive performance in pwMS. Future studies are needed to determine whether higher training dose are beneficial.