Limited evidence is available on the association between temperature and years of life lost (YLL). We applied distributed lag non-linear model to assess the nonlinear and delayed effects of temperature on YLL due to cause-/age-/education-specific mortality in Guangzhou, China. We found that hot effects appeared immediately, while cold effects were more delayed and lasted for 14 days. On average, 1 °C decrease from 25 th to 1 st percentile of temperature was associated with an increase of 31.15 (95%CI: 20.57, 41.74), 12.86 (8.05, 17.68) and 6.64 (3.68, 9.61) YLL along lag 0-14 days for non-accidental, cardiovascular and respiratory diseases, respectively. The corresponding estimate of cumulative hot effects (1 °C increase from 75 th to 99 th percentile of temperature) was 12.71 (-2.80, 28.23), 4.81 (-2.25, 11.88) and 2.81 (-1.54, 7.16). Effect estimates of cold and hot temperatures-related YLL were higher in people aged up to 75 years and persons with low education level than the elderly and those with high education level, respectively. The mortality risks associated with cold and hot temperatures were greater on the elderly and persons with low education level. This study highlights that YLL provides a complementary method for assessing the death burden of temperature.