Responses to threat occur via two known independent processing routes. We propose that early, reflexive processing is predominantly tuned to the detection of congruent combinations of facial cues that signal threat, whereas later, reflective processing is predominantly tuned to incongruent combinations of threat. To test this prediction, we examined responses to threat-gaze expression pairs (anger versus fear expression by direct versus averted gaze). We report on two functional magnetic resonance imaging (fMRI) studies, one employing prolonged presentations (2 s) of threat-gaze pairs to allow for reflective processing (Study 1), and one employing severely restricted (33 ms), backward masked presentations of threat-gaze pairs to isolate reflexive neural responding (Study 2). Our findings offer initial support for the conclusion that early, reflexive responses to threat are predominantly tuned to congruent threat-gaze pairings, whereas later reflective responses are predominantly tuned to ambiguous threat-gaze pairings. These findings highlight a distinct dual function in threat perception. (Contains 2 figures and 2 tables.)