Exposure to unusually hot or cold temperatures is associated with an increased risk of mortality. Given the rapidly changing temperature distribution in France and the implementation of policies to mitigate the effects of extreme temperatures, it is important to investigate whether the effect of temperature on mortality has changed over time. A multicenter time-series analysis was conducted for 18 regions in metropolitan France over the period 1970–2015. The temporal evolution of the temperature-mortality relationship was estimated using nonlinear distributed-lag models by introducing an interaction between temperature and time. These models allow for the investigation of possible changes in the shape of the relationship, the relative risks (RR) for different temperatures, and the proportion of mortality attributable to cold and hot temperatures, with a particular focus on the most extreme events (temperatures below the 10th percentile or above the 90th percentile for the 1970–2015 period), which are most likely to be influenced by both climate change and preventive measures. The results suggest a decrease in RRs associated with extreme heat (corresponding to temperature percentiles ≥ 99, calculated over 1970–2015) since the 1970s, and an increase in RRs associated with extreme cold. For example, the RR corresponding to an increase in temperature from the 50th percentile to the 99.9th percentile for the 1970–2015 period was 2.33 [1.95:2.79] in 1975 and 1.33 [1.14:1.55] in 2015, which appears to indicate acclimatization to heat (which may result from population adaptation, socioeconomic and medical improvements, and/or the effectiveness of preventive measures implemented starting in 2004). However, this downward trend is not observed for lower temperatures. Changes in RR are gradual over time, with no abrupt change starting in 2004. Furthermore, RRs associated with unusually high temperatures for a given year compared to previous years (corresponding to ≥90th percentiles calculated over the current year and the four preceding years) have been increasing since the 1970s. The proportion of mortality attributable to the coldest temperatures (< 10th percentile) has remained stable over time, accounting for 0.63 [0.62:0.65]% of total mortality in the 1970s and 0.58 [0.56:0.60]% in the 2010s. The proportion attributable to the hottest temperatures (> 90th percentile) has increased, rising from 0.11 [0.08:0.13]% of mortality in the 1970s to 0.23 [0.21:0.24]% in the 2010s. The attributable fractions reflect the intersection of a risk level and an exposure frequency. For cold weather, low RR values but applied to a large number of days lead to a significant impact. For heat, very high RR values correspond to few days, resulting in a smaller impact than for cold weather. However, the increase in the number of hot days has led to an increase in the fraction attributable to heat in recent years, despite a decrease in RR values. The observed trends suggest a possible acclimatization of the population to a new climate and/or improvements in socioeconomic and medical conditions. Risk decreases when a temperature that was once unusual becomes more frequent (and conversely when a temperature becomes unusual). Nevertheless, the risk associated with unusual temperatures remains. Furthermore, the decrease in heat-related RR does not offset the increase in the number of hot days, and thus an increase in the impacts of heat is observed. These results underscore the importance of continuing efforts to reduce the impact of cold and hot temperatures on mortality and of taking action to mitigate climate change and thereby avoid the most adverse scenarios. In terms of adaptation, action must be taken to improve the population’s health and enhance its thermoregulatory capacity, to modify the environment to reduce exposure to the most extreme temperatures, and to strengthen the immediate response to these extreme events.