Previous evidence on the validity of the compensating differences theory has been ambiguous. This is mainly attributed to that, in most contexts, important components of worker skills are unobserved, leading to biased estimates of compensating differences. This article uses data on professional basketball players, which contain rich measures of worker ability, measures of employer nonpecuniary characteristics, and location amenities, to produce a new test of the theory. Empirical results strongly support the theory's predictions in this context. Using this data, we also find that when important measures of player skills are omitted from the specification, there is only limited evidence in support of compensating differences. Our findings indicate that in the presence of unobserved heterogeneity, the quality of the empirical results is distorted and inference on the validity of the theory is misleading.