By using the meteorological observation data from 91 weather stations in Jiangxi province during 1981-2014 and the late rice yield data of the corresponding time period, and combining physiological characteristics of late rice and climatic features of Jiangxi, the suitability models for temperature, precipitation and sunshine in the Jiangxi area are constructed. Moreover, according to the correlation between rice yield and suitability, the climate suitability index is determined. Then based on the index, the dynamic prediction model of the late rice in Jiangxi is constructed and tested so as to diagnose the climate suitability of the late rice in Jiangxi and to do dynamic prediction of the late rice yield. The results indicate that the method has high prediction accuracy of the bumper or poor harvest trend and the actual yield in the model test and forecasting test, and it could satisfy the needs of operational services.