Adam Smith's idea that wage differences reveal preferences for risk rests on strong theoretical foundations. This paper argues, however, that the dominant approach to identify compensating wage differentials?regressing individual wages on aggregate measures of risk?may lead to arbitrary estimates of these risk differentials. In a dataset with information on both, the incidence of illnesses or injuries across firms and industries, I calculate an implicit value of one injury or illness of about (1990) USD 18,800 pursuing the dominant approach. In contrast, regressing wages on the incidence of risk across firms produces a value of one injury or illness of about USD 11,300.