MPIDR Working Paper WP-2005-020The aim of our paper is to provide an answer to the questions if and why social differences in health and mortality decrease with age. Most research confirms this decrease but the reasons for it and the role of unobserved heterogeneity are unknown. The data used for our analysis come from the US Health and Retirement Study (n=9376) and from the Danish Demographic Database (Denmark’s population above age 58). They offer detailed information about SES and health information. The technique of event-history-analysis is used, and frailty models address mortality selection. A new method is developed to consider systematic difference in the change of average frailty over age between social groups. SES differentials in mortality converge with age in Denmark but not in the US. In both countries, they converge strongly with decreasing health. When controlled for health, the differences are stable across age in both countries. This means that worsening health levels social mortality differences and not increasing age. Controlling for mortality selection removes the converging pattern over age.