Abstract: In order to study the causes of coal mine roof accident and prevent the occurrence of roof accident, based on the roof accident investigation report, the cause variables of coal mine roof accident were selected from four aspects of human, equipment, environment and management. Through the correlation among variables, Bayesian network software GeNie was used to build the Bayesian network model (BN) of the cause of coal mine roof accident, and the accuracy of Bayesian network model was tested by cross validation method, the structure and parameters of the model were studied, and the conditional probability distribution and posteriori probability of each node were calculated. Finally, the variable sensitivity and maximum cause chain analysis were explored to find out the path of key factors of accident, which was helpful to reduce the incidence of accident. The results show that in terms of human factors, the proportion of failing to fulfill the operation regulations and inadequate supervision and inspection is high, and the probability values are all greater than 84%; in terms of management factors, the confusion of safety management is the main inducement leading to accidents, and the probability value is greater than 95%; roof collapse, support problems, and whether to enter the caving area are the important factors leading to roof accidents.