The value of a statistical life (VSL) is a key component in many regulatory analyses. Along with stated preference and human capital procedures, revealed preference techniques based on markets for labor or safety-related products are commonplace in determining its magnitude.  Wage-hedonics uses the fact that workers who select riskier occupations will be compensated with a higher wage rate.  However, according to occupational sorting theory (Roy 1951), observed wage distributions are distorted by individuals selecting jobs according to both common and idiosyncratic returns.  We show that this type of sorting will typically lead to a bias in wage-hedonic VSL estimates, and we demonstrate two simple approaches to estimation that correct for it.  Implementing these strategies with data from the Current Population Survey, we recover VSL estimates that are two to three times larger than those based on the traditional wage-hedonic model, statistically significant, and robust to a wide array of specifications.