Background: Supervised exercise training alleviates motor symptoms in people with Parkinson's disease (PD). However, the efficacy of exercise to improve nonmotor symptoms such as cognitive function is less well known. Objective: To systematically review evidence on the efficacy of different exercise modes (coordination exercise, resistance exercise, aerobic exercise) on domain-specific cognitive function in patients with PD. Methods: Parallel-group randomized controlled trials published before March 2018 were included. Primary outcome measures included global cognitive function and its subdomains, and the Unified Parkinson's Disease Rating Scale was included as a secondary outcome. Methodological quality was assessed using the Physiotherapy Evidence Database scale. Results: The literature search yielded 2,000 articles, of which 11 met inclusion criteria. 508 patients (mean age 68 +/- 4 years) were included with a disease severity from 1 to 4 on the Hoehn & Yahr stage scale. Overall study quality was modest (mean 6 +/- 2, range 3-8/10). In 5 trials a significant between-group effect size (ES) was identified for tests of specific cognitive domains, including a positive effect of aerobic exercise on memory (ES = 2.42) and executive function (ES = 1.54), and of combined resistance and coordination exercise on global cognitive function (ES = 1.54). Two trials found a significant ES for coordination exercise (ES = 0.84-1.88), which led to improved executive function compared with that of non-exercising control subjects. Conclusion: All modes of exercise are associated with improved cognitive function in individuals with PD. Aerobic exercise tended to best improve memory; however, a clear effect of exercise mode was not identified.