Abstract: Based on many measured data, the limitation of empirical formula was studied. Furthermore, boundedness of BP neural networks has discussed in the tier of theory. The superiority of using support vector machine to forecast development height of water flowing fractured zone is discussed, According to 43 collected roof fracture development height measured data, in accordance with the principles of getting simple parameters and having relatively high correlation with development height of water flowing fractured zone, taking mining height, roof overburden strength and mining methods as characteristic factor, the article establishes a high degree regression model for development height of water flowing fractured zone. The model is used to predict development height of water flowing fractured zone in Xieqiao 1121 and Baizhuang 7507 working face. The results show that the use of SVM regression models to predict the development height of water flowing fractured zone has good feasibility.