Abstract Background Post-mortem studies have demonstrated considerable dendritic pathologies among persons with schizophrenia and to some extent among those with bipolar-I disorder(BD-I). Modeling gray matter(GM) microstructural properties is now possible with a recently proposed diffusion-weighted MRI modeling technique: Neurite-Orientation Dispersion and Density Imaging(NODDI). This technique may bridge the gap between neuroimaging and histopathological findings. Methods We performed an extended series of multi-shell diffusion-weighted imaging and other structural imaging series using a 3T MRI. Participants scanned included individuals with schizophrenia(n=36), BD-I(n=29), and healthy controls(HC;n=35). GM-based spatial statistics was used to compare NODDI-driven microstructural measures(orientation-dispersion-index and neurite-density-indexNDI) among groups, and assess their relationship with neurocognitive performance. We also investigated the accuracy of these measures in the prediction of group membership, and whether combining them with cortical thickness and white matter fractional anisotropy further improved accuracy. Results The GM-NDI was significantly lower in temporal pole, anterior parahippocampal gyrus, and hippocampus of the schizophrenia patients than the HC. The GM-NDI of BD-I patients did not differ significantly from either schizophrenia patients or HC, and was intermediate between the two groups in the post-hoc analysis. Irrespective of diagnosis, higher performance in spatial working memory was significantly associated with higher GM-NDI mainly in the fronto-temporal areas. Addition of GM-NDI to cortical thickness resulted in higher accuracy to predict group membership. Conclusions GM-NDI captures brain differences in the major psychoses not accessible with other structural-MRI methods. Given the strong association of GM-NDI with disease state and neurocognitive performance, its potential utility for biological subtyping should be further explored.