Abstract: To overcome shortcomings of slow convergence speed and falling into local minimum easily, traditional BP neural network was improved by using VLBP algorithm based on adaptive learning rate and LMBP algorithm based on numerical optimization technique in Matlab neural network toolbox, and a system of coal and gas outburst prediction based on the improved BP neural network was designed. Both traditional BP neural network and the improved one were used to predict coal and gas outburst experimentally. The results show that the improved BP neural network can predict coal and gas outburst faster and more accurately.