Although abnormalities in emotion recognition during a depressed episode have frequently been reported in patients with depression, less is known about the stability of these abnormalities. To examine the stability of emotion recognition abnormalities, this longitudinal study assessed patients with unipolar depression on three separate occasions at 3-monthly intervals. Recognition of sad, angry, fearful, disgusted, happy and neutral facial expressions was assessed in a matching task and a labelling task. Patients performed as well as matched healthy controls on the matching task. On the labelling task, patients showed higher accuracy and higher response bias than controls for sad expressions only, which remained stable over a 6-month interval. Over the same period, symptom severity, as measured with the Beck Depression Inventory and the Hamilton Depression Rating Scale, decreased significantly in the patient group. Furthermore, labelling performance for sad expressions was not associated with symptom severity or with changes in severity over time. This stable bias for sad expressions might signal a vulnerability factor for depression, as proposed by cognitive theories of depression. (C) 2009 Elsevier Ireland Ltd. All rights reserved.