With the trend toward aging, increases in health care expenditures are expected. Insight into (future) needs for care services requires a taxonomy of older persons' health conditions: how health status develops as people age and how these health conditions determine residual life expectancy. In this paper we provide this information for the Netherlands. We apply a flexible nonparametric method--the Grade of Membership method--to a national database and summarize the multidimensional concept of health status into a limited set of interpretable indices. We then use these indices in our panel data model for health status and mortality. The model results are used to calculate age-health profiles and expected residual lifetimes in specific health states.