Abstract: In order to overcome the disadvantages of the traditional test method of frictional resistance coefficient,such as the large amount of test and low test efficiency,the main factors influencing the frictional coefficient can be concluded based on the theory of frictional resistance coefficient and the actual data,the main factors are roadway sectional area,roadway perimeter,roadway support and roadway sectional shape. We constructed the prediction model of frictional resistance coefficient based on BP neural network, selecting the typical data as learning and test samples of BP neural network, then got on the network training by using Matlab software until getting the optimized network model.The random measurement points of Banshi mine and Daming Ⅰ mine can be predicted by using the optimized network model. The error between predicted values and measured values is no more than 10 percent. It proved that the prediction results of the network model have high reliability and engineering practice value.