Processing of facial expressions of emotion is central to human interaction, and has important effects on behaviour and affective state. A range of methods and paradigms have been used to investigate various aspects of abnormal processing of facial expressions in major depression, including emotion specific deficits in recognition accuracy, response biases and attentional biases. The aim of this review is to examine and interpret data from studies of facial emotion processing in major depression, in the context of current knowledge about the neural correlates of facial expression processing of primary emotions. The review also discusses the methodologies used to examine facial expression processing. Studies of facial emotion processing and facial emotion recognition were identified up to December 2009 utilizing MEDLINE and Web of Science. Although methodological variations complicate interpretation of findings, there is reasonably consistent evidence of a negative response bias towards sadness in individuals with major depression, so that positive (happy), neutral or ambiguous facial expressions tend to be evaluated as more sad or less happy compared with healthy control groups. There is also evidence of increased vigilance and selective attention towards sad expressions and away from happy expressions, but less evidence of reduced general or emotion-specific recognition accuracy. Data is complicated by the use of multiple paradigms and the heterogeneity of major depression. Future studies should address methodological problems, including variations in patient characteristics, testing paradigms and procedures, and statistical methods used to analyse findings.