Can prior expectancies shape attention to threat? To answer this question, we manipulated the expectancies of spider phobics and nonfearful controls regarding the appearance of spider and bird targets in a visual search task. We observed robust evidence for expectancy influences on attention to birds, reflected in error rates, reaction times, pupil diameter, and heart rate (HR). We found no solid effect, however, of the same expectancies on attention to spiders; only HR revealed a weak and transient impact of prior expectancies on the orientation of attention to threat. Moreover, these asymmetric effects for spiders versus birds were observed in both phobics and controls. Our results are thus consistent with the notion of a threat detection mechanism that is only partially permeable to current expectancies, thereby increasing chances of survival in situations that are mistakenly perceived as safe.