The present study aimed to clarify: 1) the presence of depression-related attention bias related to a social stressor, 2) its association with depression-related attention biases as measured under standard conditions, and 3) their association with impaired stress recovery in depression. A sample of 39 participants reporting a broad range of depression levels completed a standard eye-tracking paradigm in which they had to engage/disengage their gaze with/from emotional faces. Participants then underwent a stress induction (i.e., giving a speech), in which their eye movements to false emotional feedback were measured, and stress reactivity and recovery were assessed. Depression level was associated with longer times to engage/disengage attention with/from negative faces under standard conditions and with sustained attention to negative feedback during the speech. These depression-related biases were associated and mediated the association between depression level and self-reported stress recovery, predicting lower recovery from stress after giving the speech.