CONTEXT: Documenting trends in the health impacts of ambient temperature is key to supporting adaptation strategies to climate change. This paper explores changes in the temperature-related mortality in 18 French urban centers between 1970 and 2015. METHOD: A multicentric time-series design with time-varying distributed lag nonlinear models was adopted to model the shape of the relationship and assess temporal changes in risks and impacts. RESULTS: The general shape of the temperature-mortality relationship did not change over time, except for an increasing risk at very low percentiles and a decreasing risk at very high percentiles. The relative risk at the 99.9(th) percentile compared to the 50(th) percentile of the 1970-2015 temperature distribution decreased from 2.33 [95% confidence interval (CI): 1.95:2.79] in 1975 to 1.33 [95% CI: 1.14:1.55] in 2015. Between 1970 and 2015, 302,456 [95% CI: 292,723:311,392] deaths were attributable to non-optimal temperatures, corresponding to 5.5% [95% CI: 5.3:5.6] of total mortality. This burden decreased progressively, representing 7.2% [95% CI: 6.7:7.7] of total mortality in the 1970s to 3.4% [95% CI: 3.2:3.6] in the 2000s. However, the contribution of hot temperatures to this burden (higher than the 90(th) percentile) increased. DISCUSSION: Despite the decreasing relative risk, the fraction of mortality attributable to extreme heat increased between 1970 and 2015, thus highlighting the need for proactive adaptation.