BACKGROUND: A significant gap exists in the knowledge and procedural skills of medical graduates when it comes to managing emergencies. In response, highly immersive virtual reality (VR)-based learning environments have been developed to train clinical competencies. However, robust evidence on how VR-based methods affect both short- and long-term learning outcomes, as well as physiological and perceived stress, remains limited. OBJECTIVE: This study aimed to assess the effectiveness of VR-based simulation training, augmented with automated feedback, compared with video seminars at improving emergency medical competency among medical students. Furthermore, the study investigated the relationship between learning outcomes and physiological stress markers. The evaluation of participants' perceived stress and estimated learning success was also performed to provide a more comprehensive insight into VR's potential role in emergency training. METHODS: In total, 72 senior medical students underwent VR-based emergency training (intervention) or viewed video seminars (control) on 2 topics (acute myocardial infarction and exacerbated chronic obstructive pulmonary disease) in an intraindividual crossover design. Levels of applied knowledge were assessed objectively by open-response tests pre- and postintervention and after 30 days. In addition, 2 electrodermal activity markers representing physiological stress response were measured during VR sessions using a wearable sensor. Participants also rated their estimated learning success and perceived stress. They also completed self-ratings of perceived stress and estimated learning success. RESULTS: Short-term knowledge gains were comparable between the VR (mean 26.6%, SD 15.3%) and control (mean 27.2%, SD 16%) condition. However, VR training produced significantly higher long-term knowledge gains (VR: mean 17.8%, SD 15.1% vs control: mean 11.9%, SD 18%; difference: -5.9, 95% CI -11.5 to -0.4). Overall retention scores were likewise higher for VR (mean 75.4%, SD 12.5%) than for video-based learning (mean 69.0%, SD 14.5%), a difference that was more pronounced in the myocardial infarction scenario. Participants rated the VR format as significantly more effective (mean 4.83, SD 0.41, on a 5-point scale) than the video seminar (mean 3.44, SD 1.00). While physiological stress markers increased during VR sessions, their correlation with knowledge gains was weak and negative. No significant relationship was detectable between perceived stress and objective learning outcomes. CONCLUSIONS: VR-based simulation training with automated feedback may offer long-term learning advantages over a traditional video seminar in emergency-medicine education. Given the time constraints and resource limitations of clinical education, self-moderated VR-based learning may represent a valuable addition to conventional training methods. Future research could investigate the learning effects of VR scenarios regarding the retention of practical skills, as well as the impact of repeated or team-based scenarios.