Abstract: The height of water flowing fractured zone is the theoretical basis and key parameter of water-preserved mining in western mining areas of China. In recent years, BP neural network has been widely used to predict the height of water flowing fracture zone, but it has such defects as slow convergence speed and a tendency to fall into local minimum. In order to improve the prediction accuracy of the height of water flowing fractured zone, the weight values and thresholds of BP neural network were optimized by particle swarm optimization(PSO), and a prediction model was established based on PSO-BP neural network. Mining thickness, mining depth, inclined length of working face, dip angle of coal seam, overburden structural characteristics were chosen as the main influential factors of the height of water flowing fractured zone, and 22 measured data of the height of water flowing fractured zone were selected to train PSO-BP neural network. Then the trained PSO-BP neural network was used to predict two test samples, and the results were compared with the actual values, and with the predicting results of BP neural network prediction model and empirical formulas. The research results show that the average relative error of PSO-BP neural network prediction model is 1.55%, and that of BP neural network prediction model and the minimum relative error of empirical formulas are 4.8% and 9.4% respectively. The prediction accuracy of PSO-BP neural network is obviously significantly better than BP neural network and empirical formulas, and the variation of its absolute error and relative error are relatively stable, so PSO-BP neural network can effectively predict the height of water flowing fractured zone.