Distinguishing bipolar disorder from major depressive disorder is a major challenge in psychiatric treatment. Consequently, there has been growing interest in identifying neuronal biomarkers of disorder-specific pathophysiological processes to differentiate affective disorders. Thirty-six depressed bipolar patients, 36 depressed unipolar patients, and 36 matched healthy controls (HCs) participated in an fMRI experiment. Emotional faces served as stimuli in a matching task. We investigated neural activation towards angry, fearful, and happy faces focusing on prototypical regions related to emotion processing, ie, the amygdala and the anterior cingulate gyrus (ACG). Furthermore, we employed a whole-brain and a multivariate pattern classification analysis. Unipolar patients showed abnormally reduced ACG activation toward happy and fearful faces compared with bipolar patients and HCs respectively. Furthermore, the whole-brain analysis revealed significantly increased activation in bipolar patients compared with unipolar patients in the fearful condition in the right frontal and parietal cortex. Moreover, the multivariate pattern classification analysis yielded significant classification rates of up to 72% based on ACG activation elicited by fearful faces. Our results question the rather 'amygdalocentric' neurobiological models of mood disorders. We observed patterns of abnormally reduced ventral and supragenual ACG activation, potentially indicating impaired bottom-up emotion processing and automatic emotion regulation specifically in unipolar but not in bipolar individuals.