The research on the application of artificial neural networks for calamitous (rainstorm) weather forecast is reported. The mathematical description of weather forecast,the difficult points of the problem,the principle of neural network for forecast,the characteristics of rainstorm forecast and our improvements of neural network models are described and analyzed. The results show that by learning from samples, the networks can extract enough classified imformations from original data and reach good forecast index. It means further works in this direction are worth.