Learning to avoid threats often occurs by observing others. Most previous research on observational fear learning (OFL) in humans has used pre-recorded standardized video of an actor and thus lacked ecological validity. Here, we aimed to enhance ecological validity of the OFL by engaging participants in a real-time observational procedure (35 pairs of healthy male friends, age 18–27). One of the participants watched the other undergo a differential fear conditioning task, in which a conditioned stimulus (CS+) was paired with an aversive electric shock and another stimulus (CS−) was always safe. Subsequently, the CS+ and CS− were presented to the observer to test the OFL. While the friend’s reactions to the shock elicited strong skin conductance responses (SCR) in all observers, subsequent differential SCRs (CS+ > CS−) were found only when declarative knowledge of the CS+/US contingency (rated by the participants) was acquired. Contingency-aware observers also showed elevated fear potentiated startle responses during both CS+ and CS− compared to baseline. We conclude that our real-time procedure can be effectively used to study OFL. The procedure allowed for dissecting two components of the OFL: an automatic emotional reaction to the response of the demonstrator and learning about stimulus contingency.