To realize online monitoring and safety assessment of in-wheel motor running state for electric vehicles, an online method for mechanical fault diagnosis of in-wheel motor based on dynamic Bayesian networks (DBNs)is proposed in this paper. Multiple parameters highly sensitive to common mechanical faults are chosen as the characterization of in-wheel motor running state and observed nodes of DBNs under the consideration of the effects of vehicle speed on vibration signals of in-wheel motor in time domain and frequency domain for the safety of in-wheel motor running state. To solve the problem that the state transition probability distribution cannot be constructed between two continuous “time slices” of in-wheel motor running state, the mechanical fault diagnosis models for online mechanical fault diagnosis can be established based on two continuous “speed slices” with the state transition probability distributions between the different speeds. Finally, the effectiveness of this method is verified by the experiments of in-wheel motor test bench.