From a paper in the current issue of Nature:
The authors describe a patient (SM) who has bilateral brain lesions in the amygdala, a region of the medial temporal lobe known to be critical for the perception of fear. SM cannot recognize fear from facial expressions, and Adolphs et al. show that this is because she fails to look spontaneously towards the eyes on a face. When shown a face displaying an unmistakable expression of terror, she tends to fixate unworriedly on the nose and mouth regions, neglecting to notice the wide, scared eyes. Thus, she erroneously judges that the face has a neutral expression. By contrast, normal people always look immediately at the eye region of a face, and all the more so when the face is fearful.
SM avoids the eyes of all faces, no matter what their expression. But, remarkably, only her perception of fear is impaired â€” she can recognize other emotions. This suggests that visual cues provided by the eyes are particularly critical for the recognition of fear; other facial emotions can presumably be recognized without looking at the eyes (happiness can be inferred from a smile, for example).