Objective To explore the relationship between temperature and non-accidental mortality in Guangzhou, Changsha and Kunming;to evaluate the temperature-related risk of mortality; and thereby to provide scientific evidence for enacting the policy to tackle climate changes. Method Daily meteorology data and mortality data were collected in 2006-2009 in Guangzhou,Changsha and Kunming. Distributed lag non-linear model (DLNM) was established and applied in a case-crossover design, which controlled the secular trend of time, to estimate the specified effects of temperature on non-accidental mortality at conditions of lag 0-2,lag 0-18 and lag 0-27 days, respectively. Result An obvious seasonal periodicity was found in non-accidential mortality in Guangzhou, Changsha and Kunming during 2006-2009. The mortality number was comparatively high in the winters,and some high temperature days in summer; but was comparatively low in springs and autumns. An L-shaped relationship was found between temperature and mortality in Guangzhou and Kunming and a U-shaped relationship was found in Changsha. When daily mean temperature exceeded 28. 2 ℃,24. 5℃ and 23. 2 ℃, as average temperature increase 1℃,non-accidental mortality increased 4.56% (95% CI:2.74%-6. 63%),5.66% (95% CI:0. 22%-12. 65%),- 3. 94% (95% CI: -32. 77%-39. 01 %),respectively ; when daily mean temperature below 24. 8 ℃, 20. 0 ℃ and 17. 3℃, as average temperature decrease 1 ℃,the corresponding increase in non-accidental mortality were 3. 28% (95%CI:2.41%4. 10%) (lag 0-18 days),1. 35% (95% CI:0. 31%-1. 77%) (lag 0-2 days) and 2, 42% (95% CI:1. 08%-3. 27%) (lag 0-27 days),respectively. The effects of hot weather were acute and short term; while the effects of cold weather had a several days delay, but a longer persistence. Conclusions Extreme cold and hot temperature could increase the risk of non-accidental mortality in Guangzhou, Changsha and Kunming. The effects of cold weather had a several days delay, but a longer persistence.