Despite much research, bipolar depression remains poorly understood, with no clinically useful biomarkers for its diagnosis. The paralimbic system has become a target for biomarker research, with paralimbic structural connectivity commonly reported to distinguish bipolar patients from controls in tractography‐based diffusion MRI studies, despite inconsistent findings in voxel‐based studies. The purpose of this analysis was to validate existing findings with traditional diffusion MRI metrics and investigate the utility of a novel diffusion MRI metric, entropy of diffusion, in the search for bipolar depression biomarkers. We performed group‐level analysis on 9 un‐medicated (6 medication‐naïve; 3 medication‐free for at least 33 days) bipolar patients in a major depressive episode and 9 matched healthy controls to compare: (1) average mean diffusivity (MD) and fractional anisotropy (FA) and; (2) MD and FA histogram entropy—a statistical measure of distribution homogeneity—in the amygdala, hippocampus, orbitofrontal cortex and temporal pole. We also conducted classification analyses with leave‐one‐out and separate testing dataset (N = 11) approaches. We did not observe statistically significant differences in average MD or FA between the groups in any region. However, in the temporal pole, we observed significantly lower MD entropy in bipolar patients; this finding suggests a regional difference in MD distributions in the absence of an average difference. This metric allowed us to accurately characterize bipolar patients from controls in leave‐one‐out (accuracy = 83%) and prediction (accuracy = 73%) analyses. This novel application of diffusion MRI yielded not only an interesting separation between bipolar patients and healthy controls, but also accurately classified bipolar patients from controls.