Objective: This nonrandomized trial evaluated whether classroom-based training in a smartphone-based virtual reality (VR) pedestrian environment (a) teaches schoolchildren to cross streets safely, and (b) increases their self-efficacy for street-crossing. Methods: Fifty-six children, aged 8-10 years, attending primary school in Changsha, China participated. Baseline pedestrian safety assessment occurred in the VR environment and through unobtrusive observation of a subsample crossing a street for 11 days outside school. Self-efficacy was assessed through both self-report and observation. Following baseline, children engaged in the VR for 12 days in their classrooms, honing complex cognitive-perceptual skills required to engage safely in traffic. Follow-up assessment replicated baseline. Results: Probability of crash in the VR decreased posttraining (0.40 vs. 0.09), and observational data found the odds of looking at oncoming traffic while crossing the first lane of traffic increased (odds ratio [OR] = 2.4). Self-efficacy increases occurred in self-report (proportional OR = 4.7 crossing busy streets) and observation of following crossing-guard signals (OR = 0.2, first lane). Conclusions: Pedestrian safety training via smartphone-based VR provides children the repeated practice needed to learn the complex skills required to cross streets safely, and also helps them improve self-efficacy to cross streets. Given rapid motorization and global smartphone penetration, plus epidemiological findings that about 75,000 children die annually worldwide in pedestrian crashes, smartphone-based VR could supplement existing policy and prevention efforts to improve global child pedestrian safety.