Abstract: In order to solve the problem of prediction and early warning about mine roof disasters, the methods of theoretical analysis, mechanical modeling, numerical analysis and engineering practice are comprehensively applied to analyze the roof disaster forms and mechanical mechanism of the stope. We designed the software of roof disaster early warning system which is based on multi-factor radial basis neural network mathematical model, and then prejudged the settlement of the roof rock of the stope and the degree of the disaster. The system has effectively prevented the mine roof disaster through the field practice verification.