The threat capture hypothesis states that threatening stimuli are automatically processed with higher priority than non-threatening stimuli, irrespective of observer intentions or focus of attention. We evaluated the threat capture hypothesis with respect to the early perceptual stages of face processing. We focused on an electrophysiological marker of face processing (the lateralized N170) in response to neutral, happy, and angry facial expressions displayed in competition with a non-face stimulus (a house). We evaluated how effects of facial expression on the lateralized N170 were modulated by task demands. In the pixel task, participants were required to identify the gender of the face, which made the face task-relevant and entailed structural encoding of the face stimulus. In the pixel task, participants identified the location of a missing pixel in the fixation cross, which made the face task-irrelevant and placed it outside the focus of attention. When faces were relevant, the lateralized N170 to angry faces was enhanced compared to happy and neutral faces. When faces were irrelevant, facial expression had no effect. These results reveal the critical role of task demands on the preference for threatening faces, indicating that top-down, voluntary processing modulates the prioritization of threat.