Acute depression is associated with impaired self-referential processing. Antidepressant effects on the neural bases of self-referential processing in depression are unknown. This study aimed to assess short-and long-term effects of agomelatine on these neural bases in depressed patients and the association between pre-treatment brain activation and remission of depression 6 months later. We conducted a randomized double-blind, placebo-controlled, functional magnetic resonance imaging (fMRI) study during an emotional self-referential task, including three scanning sessions (baseline, after 1 week, and after 7 weeks). Twenty-2 five depressed outpatients were included, all treated with agomelatine or placebo for 1 week. Then, all patients received agomelatine for 24 weeks. Fourteen matched healthy volunteers (HV) who received placebo for 1 week were also included. After 7 days, only depressed patients receiving agomelatine significantly deactivated the ventrolateral prefrontal cortex during self-referential processing, as observed in HV at baseline. After 7 weeks, depressed patients significantly increased the activation of the ventral anterior cingulate cortex. Finally dorsomedial prefrontal cortex and precuneus activations at baseline significantly separated remitters from non-remitters at 24 weeks. In depressed patients, agomelatine had short-and long-term effects on brain structures involved in anhedonia and emotional regulation during self-referential processing. Activation of the dorsomedial prefrontal cortex and precuneus could be informative in the development of biomarker-based treatment of major depression.