Abstract
We cannot help but impute emotions to the behaviors of others, and constantly infer not only what others are feeling, but also why they feel that way. The comprehension of other people's emotional states is computationally complex and difficult, requiring the flexible, context-sensitive deployment of cognitive operations that encompass rapid orienting to, and recognition of, emotionally salient cues; classification of emotions into culturally-learned categories; and using an abstract theory of mind to reason about what caused the emotion, what future actions the person might be planning, and what we should do next in response. This review summarizes what neuroscience data - primarily functional neuroimaging data - has so far taught us about the cognitive architecture enabling emotion understanding in its various forms.