In human beings, experiments investigating fear conditioning with social stimuli are rare. The current study aims at translating an animal model for social fear conditioning (SFC) to a human sample using an operant SFC paradigm in virtual reality. Forty participants actively (using a joystick) approached virtual male agents that served as conditioned stimuli (CS). During the acquisition phase, unconditioned stimuli (US), a combination of an air blast (5 bar, 10 ms) and a female scream (95 dB, 40 ms), were presented when participants reached a defined proximity to the agent with a contingency of 75% for CS+ agents and never for CS agents. During the extinction and the test phases, no US was delivered. Outcome variables were pleasantness ratings and physiological reactions in heart rate (HR) and fear-potentiated startle. Additionally, the influence of social anxiety, which was measured with the Social Phobia Inventory scale, was evaluated. As expected after the acquisition phase the CS+ was rated clearly less pleasant than the CS. This difference vanished during extinction. Furthermore, the HR remained high for the CS+, while the HR for the CS was clearly lower after than before the acquisition. Furthermore, a clear difference between CS+ and CS after the acquisition indicated successful conditioning on this translational measure. Contrariwise no CS+/CS- differences were observed in the physiological variables during extinction. Importantly, at the generalization test, higher socially fearful participants rated pleasantness of all agents as low whereas the lower socially fearful participants rated pleasantness as low only for the CS+. SFC was successfully induced and extinguished confirming operant conditioning in this SFC paradigm. These findings suggest that the paradigm is suitable to expand the knowledge about the learning and unlearning of social fears. Further studies should investigate the operant mechanisms of development and treatment of social anxiety disorder.