Parkinson’s Disease (PD), one of the most common neurological disorders, has long been a challenge in public health clinical diagnosis as well as scientific understanding. Recently, there has been an upsurge of interest in brain network analysis which benefits the understanding of brain functions and early detection of neurological disorders extensively. Multi-view brain networks with different connectivity patterns among regions of interests (ROIs) can be constructed to reflect different and complementary perspectives of the brain connectivity profile. However, the extraction of such multi-view brain networks relies on the availability of multiple neuroimaging modalities and heavy data preprocessing, which often leads to severe missing data in either view. The cross-view missing issue hinders the pragmaticality of multi-view representation learning and downstream analysis. In this work, we formulate the novel problem of cross-view brain network generation and propose CroGen, a graph generative model that can generate the missing view when only one view is given. Specifically, GroGen leverages the potential correlation between diverse views of brain networks of the same individuals. Moreover, we design a pre-train schema to make full use of the labeled individuals with only single views of brain networks. Extensive experiments on real-life Parkinson’s Progression Markers Initiative (PPMI) cohort demonstrate the supreme effectiveness of CroGen over baselines on both cross-view generation tasks and downstream PD classification.