The ability to navigate complex social situations is central for human interaction. Yet, effective trainings in (psychologically) safe environments for the acquisition of widely applicable social skills remain elusive. We suggest enhancing individuals' social skills and their antecedents through virtual reality (VR) to address this gap. Our research seeks to conceptualize, apply, and empirically evaluate a novel framework for social skill acquisition in immersive environments. Using a specifically developed job-interview training, we employ a preregistered 2 × 3 mixed intervention design (N = 114), comparing VR-based vs. chat-based trainings across three measurement waves (before vs. immediately after vs. four months after training). Results show that our VR intervention significantly improved individuals' job-interview self-efficacy and lowered their task-related anxiety. The chat-based intervention had similarly favorable effects. The positive effects on self-efficacy and anxiety from both trainings persisted over four months. The VR training, however, required only 50% of the training time for comparable success, and participants reported a preference for the VR experience—recalling it more vividly and expressing a higher willingness to reengage in similar trainings. Mediation results align with our proposed framework for social skill acquisition in immersive environments and establish VR's effectiveness through key psychological pathways—physiological arousal, as well as cognitive and motivational factors. Our work contributes a conceptual framework to the growing body of literature on learning in VR and empirically highlights that VR trainings can provide an engaging and efficient method for training complex social behaviors in a simulated, safe, standardized, and scalable environment.