With 0800 h convenient data and numerical products as input,the element field of a key region is expanded as the Chebyshev orthogonal polynomial,whose coefficients are taken as predictors to be trained with the aid of the B-P neural network,whereupon is established an artificial rainfall fallout region forecasting network for categories 5,10 and 25 mm for 24 h in advance.Tests show that >95% of the historical events are predicted with successful operational application in summer(JJA)1994.