Schizophrenia (SZ) and bipolar disorder (BD) are two of the most frequent mental disorders. These disorders exhibit similar psychotic symptoms, making diagnosis challenging and leading to misdiagnosis. Yet, the network complexity changes driving spontaneous brain activity in SZ and BD patients are still unknown. Functional entropy (FE) is a novel way of measuring the dispersion (or spread) of functional connectivities inside the brain. The FE was utilized in this study to examine the network complexity of the resting-state fMRI data of SZ and BD patients at three levels, including global, modules, and nodes. At three levels, the FE of SZ and BD patients was considerably lower than that of normal control (NC). At the intra-module level, the FE of SZ was substantially higher than that of BD in the cingulo-opercular network. Moreover, a strong negative association between FE and clinical measures was discovered in patient groups. Finally, we classified using the FE features and attained an accuracy of 66.7% (BD vs. SZ vs. NC) and an accuracy of 75.0% (SZ vs. BD). These findings proposed that network connectivity’s complexity analyses using FE can provide important insights for the diagnosis of mental illness.