Abstract: The surface subsidence coefficient is an important parameter in the prediction of surface subsidence. The accuracy of its value will have a direct impact on the prediction results of subsidence. Because there are many factors affecting the surface subsidence coefficient of coal mining，and there are complex relationships such as uncertainty and nonlinearity among the factors，the prediction of surface subsidence coefficient is very difficult. In order to solve the problem that it is difficult to accurately predict the surface subsidence coefficient and improve the prediction accuracy，a prediction model of surface subsidence coefficient is established based on the measured surface movement observation data of 35 mining areas in China. The combination weight of influencing factors of surface subsidence coefficient is obtained by combining grey correlation analysis and principal component analysis. Eight influencing factors such as mining thickness，coal seam dip angle，average mining depth，strike width depth ratio，dip width depth ratio，advancing speed，loose layer thickness and average firmness coefficient of overburden are selected. According to the combination weight，the influencing factors of surface subsidence coefficient in the data of surface movement observation station are sorted，and the main influencing factors of surface subsidence coefficient are obtained，and the main influencing factors are taken as the input and the surface subsidence coefficient is taken as the input parameter. Then a BP neural network model for the prediction and analysis of surface subsidence coefficient is proposed. The results show that:the combination weight of loose layer thickness，advancing speed，average mining depth and dip width depth ratio is larger，which is the main influencing factor of surface subsidence coefficient; the BP neural network prediction model of surface subsidence coefficient established by the main influencing factors of surface subsidence coefficient has high prediction accuracy，the minimum absolute error is 3.954%，the maximum value is only - 6.918%，and the average relative error can reach 7.179%，which is very close to the measured value. The prediction accuracy of the model can meet the basic engineering needs，and it is a feasible method to accurately predict the surface subsidence coefficient.