Adam Smith's idea that wage differences reveal preferences for risk rests on firm theoretical foundations. This paper argues, however, that the standard approach to identify these differentials in practice may be flawed. Empirical practice usually identifies compensating wage differentials for risk by regressing individual wages on aggregate measures of risk, usually industry or occupation average risk. If jobs differ within industries or occupations, the ''aggregate approach'' may identify arbitrary compensating differentials for risk. In a dataset with precise information on job risk as well as aggregate risk, I demonstrate that using aggregate risk identifies wage differentials that are two to five times larger than wage differentials based on job riskninformation. This result is robust to controlling for time constant unobserved individual or job heterogeneity.