This study investigated the impact of racial congruence between instructors and learners on learning performance within a virtual reality (VR) environment, and the mediating role of self-efficacy. Participants (N = 25) were randomly assigned to interact with either a racially congruent or incongruent virtual instructor while undergoing safety training in a virtual construction site. We hypothesized that self-efficacy would mediate the effect of racial congruence on learning performance. The study found that learners with racially congruent instructors in VR showed superior learning performance and higher self-efficacy. While self-efficacy was a significant predictor of learning, we could not prove that it acts as a mediator between racial congruence and learning performance. These findings highlight the potential of VR for investigating social dynamics in learning and emphasize the importance of representation.