In this paper we argue that in order to test competing hypotheses on the emergence of social mortality differentials, one has to adopt a long-term perspective. Studying social inequality in mortality in Geneva from 1625 to 2005, we use historical mortality data published by different authors and contemporary data drawn from an ongoing research project. The comparison over four centuries gives evidence to both the constancy and convergence hypotheses. Mortality is systematically lower-than-average among elites on the one hand, but on the other hand the difference between the top and the bottom of the social ladder is decreasing over time.