Affective facial computing (AFC) refers to the use of computer techniques to analyze and synthesize facial behavior that may be related to emotion and other cognitive-affective states. This chapter evaluates the state-of-the-art in the generalizability of AFC systems across domains and also reviews approaches that are designed to improve such generalizability. We address two questions that we regard as critical to the field: (1) To what extent do AFC systems generalize to new data sources, or domains? (2) What approaches show the most promise in improving generalizability?