Recent studies have revealed that the effect of temperature on mortality has changed over time. One of the major contributors to the changes is adaptation. We aimed to understand the relationship between elderly mortality and temperature anomaly using the temperature deviation index (TDI), which considers exposure history. Summertime (May to September) mortality data from 1996 to 2014 and meteorological data from 1971 to 2014 were obtained for 16 regions covering South Korea. The TDI was defined as the target day's temperature abnormality compared to previous 25 years' apparent temperature (AT). The relationship between the TDI and elderly mortality for each region was examined by generalized linear modeling with Poisson distribution. Pooled estimates were computed to yield a national effect estimate. Stratified analyses were performed using the percentiles of AT and TDI. Most regions showed positive linear associations, and the associations ranged from 0.4 to 4.3% increase per unit increase of the TDI. In the pooled analyses, a unit increase of the TDI was associated with a 1.4% increase (95% confidence interval [CI] 0.93-1.87) in elderly mortality. In the stratified analysis, the relationship between the TDI and elderly mortality was significant at or above the 75th percentile of AT (1.32% increase; 95% CI 0.47-2.22). We suggest a positive association between the TDI and elderly mortality in South Korea. The association observed particularly in the highest percentile of AT in the stratified analysis suggests independent effects of temperature anomaly in addition to those of absolute AT.