Objective: To conduct a systematic review to determine the efficacy of exercise-based interventions on improving performance-based measures of physical function and markers of physical frailty in community-dwelling, frail older people. Data sources: Comprehensive bibliographic searches in MEDLINE, the Cochrane Library, PEDro, and CINAHL databases were conducted (April 2013). Study selection: Randomized controlled trials of community-dwelling older adults, defined as frail according to physical function and physical difficulties in activities of daily living (ADL). Included trials had to compare an exercise intervention with a control or another exercise intervention, and assess performance-based measures of physical function such as mobility and gait, or disability in ADL. Data extraction: Two review authors independently screened the search results and performed data extraction and risk of bias assessment. Nineteen trials were included, 12 of them comparing exercise with an inactive control. Most exercise programs were multicomponent. Data synthesis: Meta-analysis was performed for the comparison of exercise versus control with the inverse variance method under the random-effects models. When compared with control interventions, exercise was shown to improve normal gait speed (mean difference [MD]=.07m/s; 95% confidence interval [CI], .04-.09), fast gait speed (MD=.08m/s; 95% CI, .02-.14), and the Short Physical Performance Battery (MD=2.18; 95% CI, 1.56-2.80). Results are inconclusive for endurance outcomes, and no consistent effect was observed on balance and the ADL functional mobility. The evidence comparing different modalities of exercise is scarce and heterogeneous. Conclusions: Exercise has some benefits in frail older people, although uncertainty still exists with regard to which exercise characteristics (type, frequency, duration) are most effective.