Objective To understand the associations between extremely low and high air temperature and the years of life lost (YLL) due to diabetes deaths in Chongqing and Harbin with different climatic characteristics in China. Methods A double threshold B-spline distributed lag non-linear model (DLNM) was used to investigate the lag and cumulative effects of extremely low and high air temperature on YLL due to diabetes for lag 0-30 days by using the urban meteorological and diabetes mortality data of Chongqing (2011-2013) and Harbin (2008-2010). The effects were expressed as relative risk (RR). Results In Chongqing, the cold effects on YLL due to diabetes were delayed by four days and lasted for three days (lag4-6) with the highest RR of 1.304 (95% CI: 1.033-1.647) at lag5. The hot effects were delayed by one day (lagl) with RR of 1.321 (95% CI: 1.061-1.646). In Harbin, the extreme cold effects on YLL were delayed by four days and lasted for seven days (lag4-10) with the highest RR of 1.309 (95%CI: 1.088-1.575) at lag6. The hot effects were delayed by one day and lasted for four days (lag 1-4) with the highest RR of 1.460 (95% CI: 1.114-1.915) at lag2. The unit risk for cold and hot effects was 43.7% (P=0.005 5) and 18.0% (P= 0.000 2) in Chongqing and 15.0% (P=0.000 8) and 29.5% (P=0.001 2) in Harbin, respectively. Conclusions Both extremely low air temperature and extremely high air temperature might increase the years of life lost due to diabetes in cities with different climate characteristics. Health education about diabetes prevention should provide information about the effects of extreme weather events.