Objective: To identify what features of social media – promotional messaging or peer networks – can increase physical activity. Method: A 13-week social media-based exercise program was conducted at a large Northeastern university in Philadelphia, PA. In a randomized controlled trial, 217 graduate students from the University were randomized to three conditions: a control condition with a basic online program for enrolling in weekly exercise classes led by instructors of the University for 13 weeks, a media condition that supplemented the basic program with weekly online promotional media messages that encourage physical activity, and a social condition that replaced the media content with an online network of four to six anonymous peers composed of other participants of the program, in which each participant was able to see their peers' progress in enrolling in classes. The primary outcome was the number of enrollments in exercise classes, and the secondary outcomes were self-reported physical activities. Data were collected in 2014. Results: Participants enrolled in 5.5 classes on average. Compared with enrollment in the control condition (mean = 4.5), promotional messages moderately increased enrollment (mean = 5.7, p = 0.08), while anonymous social networks significantly increased enrollment (mean = 6.3, p = 0.02). By the end of the program, participants in the social condition reported exercising moderately for an additional 1.6 days each week compared with the baseline, which was significantly more than an additional 0.8 days in the control condition. Conclusion: Social influence from anonymous online peers was more successful than promotional messages for improving physical activity. Clinical Trial Registration: ClinicalTrials.gov: NCT02267369.