Attention is attracted exogenously by physically salient stimuli, but this effect can be dampened by endogenous attention settings, a phenomenon called "contingent capture." Emotionally salient stimuli are also thought to exert a strong exogenous influence on attention, especially in anxious individuals, but whether and how top-down attention can ameliorate bottom-up capture by affective stimuli is currently unknown. Here, we paired a novel spatial cueing task with fMRI to investigate contingent capture as a function of the affective salience of bottom-up cues (face stimuli) and individual differences in trait anxiety. In the absence of top-down cues, exogenous stimuli validly cueing targets facilitated attention in low-anxious participants, regardless of affective salience. However, although high-anxious participants exhibited similar facilitation following neutral exogenous cues, this facilitation was completely absent following affectively negative exogenous cues. Critically, these effects were contingent on endogenous attentional settings, such that explicit top-down cues presented before the appearance of exogenous stimuli removed anxious individuals' sensitivity to affectively salient stimuli. fMRI analyses revealed a network of brain regions underlying this variability in affective contingent capture across individuals, including the fusiform face area (FFA), posterior ventrolateral frontal cortex, and SMA. Importantly, activation in the posterior ventrolateral frontal cortex and the SMA fully mediated the effects observed in FFA, demonstrating a critical role for these frontal regions in mediating attentional orienting and interference resolution processes when engaged by affectively salient stimuli.