We use data on twins matched to register-based information on earnings to examine the long-standing puzzle of non-existent compensating wage differentials. The use of twin data allows us to remove otherwise unobserved productivity differences that were the prominent reason for estimation bias in...