In this paper we exploit a unique database that merges longitudinal earnings data on Pennsylvanian workers with national death records to study the detailed nature of the correlation between earnings and mortality. We find that the estimates typically reported in the literature, which are based on single years of earnings data, are likely to understate substantially the strength of the association between income and mortality. In particular, relative to a single year of earnings, the average of earnings over a six-year period predicts a 70 percent greater impact of income on mortality. In addition, controlling for the mean level of earnings over a period, we find that greater earnings volatility is associated with higher mortality. We also examine the lag structure of the relationship between earnings and mortality. We find that conditional on a small number of years of recent earning levels, there is little or no correlation between earnings levels in earlier years and current mortality. This runs counter to the interpretation of the earnings-mortality correlation that views workers as "buying health" by allocating a steady fraction of their earnings to the accumulation of a health stock. It is also inconsistent with interpretations of the earnings- mortality link emphasizing the correlation of both variables with an unobserved, time-invariant trait, such as the rate at which workers discount the future. We find that explaining the patterns of correlation appears to require a relatively complex theory that we leave to future research.