Facial expressions offer an ecologically valid model for examining individual differences in affective decision-making. They convey an emotional signal from a social agent and provide important predictive information about one's environment (presence of potential rewards or threats). Although some expressions provide clear predictive information (angry, happy), others (surprised) are ambiguous in that they predict both positive and negative outcomes. Thus, surprised faces can delineate an individual's valence bias, or the tendency to interpret ambiguity as positive or negative. Our initial negativity hypothesis suggests that the initial response to ambiguity is negative, and that positivity relies on emotion regulation. We tested this hypothesis by comparing brain activity during explicit emotion regulation (reappraisal) and while freely viewing facial expressions, and measuring the relationship between brain activity and valence bias. Brain regions recruited during reappraisal showed greater activity for surprise in individuals with an increasingly positive valence bias. Additionally, we linked amygdala activity with an initial negativity, revealing a pattern similarity in individuals with negative bias between viewing surprised faces and maintaining negativity. Finally, these individuals failed to show normal habituation to clear negativity. These results support the initial negativity hypothesis, and are consistent with emotion research in both children and adult populations.