Abstract: Ensuring safe production is a fundamental requirement of the cropping system, being an important basis for stable and high yield of crops. Based on the daily average and maximum temperatures from 1961-2015 in 317 meteorological stations in the rice growing areas of southwestern China, the study analyzed safe sowing dates of rice under different guarantee rates (80%-95%) using the mean variance method. The earliest date of heat stress and chilling injury at rice flowering stage under different guarantee rates were analyzed too. A prediction model of the threshold temperature start date for rice at critical development stages in southwestern China was constructed using the stepwise regression method. At the 80%-95% guarantee rate, the earliest safe sowing date for rice was discovered in southern Yunnan, which was around February 20. The safe sowing dates in northern Yunnan, southwestern Sichuan, central Sichuan basin, most parts of Chongqing and parts of Guizhou were from March 10 to April 10. The safe sowing dates in the rest of southwestern China were from April 10 to May 20. At the 80%-95% guarantee rate, heat stress at flowering stage of middle-season rice occurred mainly in the Sichuan basin, Chongqing, northern and southeastern Guizhou, and the earliest occurrence dates were from July 15 to August 20. The earliest occurrence dates of chilling damage at ratooning or late rice flowering stage in most of Yunnan and Guizhou, and southwestern Sichuan were from June 20 to July 15. Meanwhile, the earliest occurrence dates of chilling damage in western Sichuan basin and northeastern Guizhou were from August 1 to August 20, and from August 20 to September 20 in the remaining areas. Based on the latitude, longitude and altitude, a model for predicting the threshold temperature initiation period for rice at critical development stages in southwestern China was established, which was simple and practical. In addition, using Hejiang of Sichuan and Meitan of Guizhou as the case study, the differences between the actual values and the simulated values of threshold temperature start date for rice were analyzed. The relative error between the actual values and the simulated values of threshold temperature start date for rice was less than 5.0%, which indicated that the model had a good simulation effect. In summary, the prediction of threshold temperature start date for rice at critical development stages can provide a theoretical basis for safe production, disaster prevention and mitigation options in rice production in southwestern China. On the one hand, according to the safe sowing period, the actual sowing period of rice can be adjusted in time, which is necessary to avoid hazards of meteorological disasters. On the other hand, the local major rice varieties can be selected through probabilistic decision-making, to ensure stable and high yield in rice production.