Central emotion theories assume that during threatening and dangerous events the human face signals a prototypical, distinct, and universally recognized expression of fear which can be accurately decoded by conspecific perceivers. Due to the importance of fear expressions, an unusually large body of research has been dedicated to exploring their evolutionary origins, neurobiological mechanisms, and clinical significance. However, these studies typically utilize highly recognizable posed actor portrayals presumed to closely resemble the diagnostic physical appearance of real-life fearful faces. Here, we challenge this diagnosticity assumption. Following context-dependent frameworks (Barrett, 2017), we hypothesized that extrafacial context (e.g., situational information, body posture, etc.) plays a far greater role in fear communication than the signal of the isolated face. In 12 preregistered experiments (N = 4,180), we examined the perception of authentic, real-life videos documenting a diverse range of intense fear-inducing situations (e.g., height jumping, physical attacks, exposure to phobia triggers). Participants viewed the face alone, the context with no face, or the full video while various response methods of emotion perception were tested (forced choice, open-ended, multiple emotion scales, valence-arousal ratings). Across experiments, videos of the faces alone failed to communicate fear in a reliable manner. In sharp contrast, context with no faces, and faces with context were clearly and robustly perceived as fearful, with medium to large effect sizes. These findings suggest that despite the undisputed importance of perceiving fear reactions, facial expressions alone bear minimal diagnostic value, while context plays a critical role in real-life fear perception.