© 2020 Society of Biological Psychiatry Background: Current diagnostic strategy for bipolar disorders relies on symptomological classification. Yet, responses to both pharmacological and psychotherapeutic treatments vary widely, suggesting that underlying neuropathological differences are not well defined by current nosology. Classifying patients with bipolar disorder based on emotion regulation network (ERN) activation may account for some of the heterogeneity within the disorder. Methods: Euthymic participants diagnosed with bipolar I disorder (n = 86) and healthy subjects (n = 80) underwent functional magnetic resonance imaging scans while engaged in emotional reappraisal of negative stimuli. After determining average regional activations in key network regions, we applied agglomerative hierarchical clustering to identify subtypes of bipolar disorder. Next, we examined relations among neural subtypes, demographic variables, and mood symptoms. Results: Analyses revealed two primary neural subtypes of euthymic bipolar I disorder participants. The first subtype, ERN cluster 1, was characterized by increased amygdala activation and slightly increased ventrolateral prefrontal and subgenual cingulate activation, whereas ERN cluster 2 was defined by decreased amygdala activation with wider-spread prefrontal activation. Cluster 1 was associated with a higher number of hospitalizations for depression (odds ratio = 1.30, 95% confidence interval = 1.02–1.64) and later onset of manic episodes (odds ratio = 1.06, 95% confidence interval = 1.00–21.13) than cluster 2. ERN clusters of healthy subjects differed from bipolar disorder clusters and were defined by differential activation of the prefrontal cortex. ERN clusters of healthy subjects, which differed from bipolar disorder clusters, were defined by differential activation of the prefrontal cortex. Conclusions: Emotion regulation circuitry can distinguish neurobiological subtypes of bipolar disorder in the euthymic state. These subtypes, which are differentially associated with indices of illness severity and subsyndromal affective symptoms, may help to inform relapse risk and more personalized treatment approaches.