A rainfall estimate system using GMS-5 data was developed through multiple regression method, which can be used to estimate six-hour rainfall hourly. Further study about its application under different precipitation situations， such as stratus, convective cloud, we found that the GMS-5 four channel, especially other derived data from them， such as infrared temperature variance with time, and the difference between infrared temperature and vapor temperature do good function to improve estimate accuracy, and the rainfall estimate function based on real-time samples pool by using step regression method can further eliminate error caused by deep cirrus or special terrain. 