There is extensive evidence that emotional--especially threatening--stimuli rapidly capture attention. These findings are often explained in terms of a hard-wired and relatively inflexible fear module. We propose an alternative, more flexible mechanism, arguing that motivational relevance is the crucial factor driving rapid attentional orienting. To test our hypothesis, we endowed initially neutral face stimuli with relevance by randomly assigning them to a social in-group or out-group during a 1-min learning phase, and used these faces as cues in a dot probe task to measure rapid attentional orienting. Across three experiments, we observed attentional orienting toward faces assigned to the out-group. Initial rapid orienting (after 100 ms, Experiments 1 and 2) was observed only for familiar faces for which group membership was explicitly encoded, suggesting that rapid orienting may be based on affectively charged memory traces. At a later time point (after 500 ms, Experiment 3), attention was deployed toward unfamiliar faces sharing a physical attribute (background color) with the familiar out-group faces, suggesting a more time-consuming on-line appraisal of the stimulus. The amount of attentional bias to out-group faces was correlated with individual differences in the accessibility of group identification. Our findings demonstrate that attentional prioritization mechanisms can be flexibly tuned by a brief learning phase of social identity. This is consistent with the idea that attention mechanisms subserving the selection and prioritization of emotional aspects of the environment are not static and hard-wired, but may rapidly adapt to recent changes in motivational contingencies. (Contains 3 tables and 3 figures.)