The 24 h rain forecasting B-P neural network of Qinghai Plateau regional rain are established with regular synoptic data and ECMWF data for 5 mm, 10 mm and 25 mm fainfall synoptic processes. The results that learning from samples show, the artificial neural network can extract enough classified informations from original data and reach good historical fitness(≥95%). The forecast experiment made in 1994 and 1995 duing June~August is better.