We present a new approach to improve the reliability of quantifying the impact of a heat wave on mortality rates. We show, for the recent European summer 2003 heat wave, that the use of absolute maximum temperature values, or number of days above a given threshold, can be misleading. Here, we have assessed the impact of the heat wave on Iberian mortality by applying a four step procedure: (1) calculating, for each observatory, the local maximum temperature (T (max)) distributions, (2) calculating the corresponding 95th percentile values (T (threshold)), (3) locally defining extremely hot days (EHD) as those days on which the local threshold of the 95th percentile of the series is exceeded, and (4) calculating the total degrees-days (DD) of exceedance, by calculating the difference T (max)-T (threshold) and summing these values for all days above T (threshold). We show that the relationship between summer mortality rates and the DD index is non-linear and can be described by a logarithmic function, with a correlation coefficient of 0.78, which explains 60.6% of the mortality variance (F value of 24.64, significant at P<0.0001). Using maximum temperatures, no significant relationship is found with mortality, whereas the EHD frequency shows a significant association with mortality, albeit weaker than that obtained with DD.