The difficulty of cascading failure analysis lies in its dynamic process which varies from the slow progression of line overloads to extremely fast cases involving system-wide instabilities as well as other various factors. In order to tackle the problem， the predictive contingency analysis method based on the Bayesian network is introduced in this paper. Based on the study of the development and characteristics of cascading failure， the power flow redistribution process after the initial disturbance is selected to be the focus of the study. With the linear probability characteristic of protection hidden failure taken into account， the cascading failure analysis model based on the Bayesian network is established for uncertainty reasoning. The method proposed is superior both in analyzing effectiveness and in model simplification to existing methods. The results of testing examples validate the correctness of the method proposed.