Rice is the main food crop in southern China. So far, low-temperature disaster has become one of the main agricultural meteorological disasters which influence the production of rice. Spring low-temperature disaster of early rice and autumn cold dew wind of late rice are the main low-temperature disasters in double-cropping rice growing areas in southern China. However, the frequency of low-temperature disaster has decreased in some regions while increased in other regions, and the damage to the rice yield even increases under the background of global warming. In order to reduce the yield loss and build comprehensive forecasting and early-warning technical architecture, the low-temperature disaster is deeply looked into. Using the software SPSS and methods of factor puffing, correlation analysis and Fisher's discriminant, a series of data are analyzed, including daily meteorological data, rice growing period data from 708 weather stations located in the planting region of double-cropping rice in the south during 1961 2010,together with meteorological industry standards. An early-warning model is established to forecast low-temperature disasters for spring rice in high risk areas (area I) 10 days in advance, and for autumn rice in both high risk areas (area I) and main disaster areas (area II) 5 days in advance. Based on data during 1961 2009,the model constructed is used for hindcast, and data of 2010 is used for evaluation. The average basically consistent accuracy of the early-warning model in area I of early rice, late japonica rice and late indica rice is 90.5%, 74.2% and 80.3%, respectively. As for area II of late japonica rice and late indica rice, the average basically consistent accuracy of the early-warning model is 89.4% and 80.3%, respectively. On the whole, the average basically consistent accuracy of the early-warning model is above 80%,and the error is within one level,showing good efficiency.