Past attempts to estimate the magnitude of compensating wage differentials have been hindered by the biasing effects of omitted variables and measurement error. We argue that a wage change formulation, estimated with panel data that contain worker reports of their own job characteristics, reduces both of these biases. Our empirical results, based on a large panel of workers in Sweden, confirm these conjectures by giving many more reasonable coefficient estimates for a wage change equation than for a wage level formulation.