In this study, a novel approach considering the dynamic interaction of train-bridge system based on the Vector Form Intrinsic Finite Element (VFIFE) method is proposed for evaluating the risk of derailment of a train traveling over a long-span cable-stayed bridge under crosswinds. Making use of the VFIFE method, the train and bridge systems can be discretized into a group of mass particles with massless stiffness elements and each mass particle satisfies the Newton's second law. The internal forces induced by pure deformations in the massless elements are calculated using the fictitious reverse motion method, by which the conventional updating, factorizing and inverse procedures for solving structural matrices of the train-bridge system are skipped. As a result, the equation of each mass particle can be solved individually. For evaluating potential derailment of a running train under crosswinds, the running safety factors including derailment factor, offload factor and lateral wheel-rail force have been obtained. In the case study, a two-phase plot with safety and derailment regions is configured for a train running on a cable-stayed bridge under crosswinds. Results show that the wind-induced vibration of the system may affect significantly the running safety of the train. When the time-varying parameters required are available from online and onboard monitoring systems, the proposed model can be executed to assess the derailment risk in a real-time manner.