This paper deals with the compensation for non-fatal accident risk in Switzerland and presents empirical estimates of the value of a statistical injury. We approach the problem of endogenous sorting of workers into jobs with different accident risks based on unobserved productivity differences twofold. First, we have access to the number of accidents not only at the level of industries, but within cells defined over industry x skill-level of the job, which allows us to estimate risk compensation within groups of workers defined over the same cells. Second, we capitalize on the partial panel structure of our data which allows us to empirically isolate the wage component specific to the employer. Our different approaches to identification in fact yield very different estimates of the value of a statistical injury. Our preferred estimate gives an estimate of about 40,000 Swiss francs (per prevented injury pernyear).