Organisms tend to respond similarly to stimuli that are perceptually close to an event that predicts adversity, a phenomenon known as fear generalization. Greater dissimilarity yields weaker behavioral responses, forming a fear-tuning profile. The perceptual model of fear generalization assumes that behavioral fear tuning results from perceptual similarities, suggesting that brain responses should also exhibit the same fear-tuning profile. Using fMRI and a circular fear-generalization procedure, we tested this prediction. In contrast with the perceptual model, insula responses showed less generalization than behavioral responses and encoded the aversive quality of the conditioned stimulus, as shown by high pattern similarity between the conditioned stimulus and the shock. Also inconsistent with the perceptual model, object-sensitive visual areas responded to ambiguity-related outcome uncertainty. Together these results indicate that fear generalization is not passively driven by perception, but is an active process integrating threat identification and ambiguity-based uncertainty to orchestrate a flexible, adaptive fear response.