In the present study, automatic processing of facial affect in clinical depression was investigated in the course of an inpatient treatment program. Patients suffering from clinical depression (n = 20) and healthy controls (n = 21) completed the facial affective priming task developed by Murphy and Zajonc (1993) twice, about 7 weeks apart. Subjects were instructed to evaluate neutral Chinese ideographs primed by masked displays of sad, happy, and neutral facial affect, including a no-prime condition. In the course of treatment, patients recovered significantly. In acutely depressed patients, no priming based on emotional faces could be found compared to neutral faces at time 1. However, compared to the no-prime condition, negative evaluation shifts elicited by neutral and sad faces were found which were significantly correlated with symptom severity. Patients with persisting high levels of depression after therapy judged ideographs more negatively in all three facial prime conditions at time 1. We conclude that clinically depressed patients are characterized by automatic processing biases for facial affect. An enhanced sensitivity for sad facial expressions and a negatively biased automatic processing of neutral and happy facial affect appears to be associated with depression persistence.