This paper proposes a new approach to estimate task prices per efficiency unit of skill in the Roy model. I show how the sorting of workers into tasks and their associated wage growth can be used to identify changes in task prices under relatively weak assumptions. The estimation exploits the fact that the returns to observable talents will change differentially over time depending on the changes in prices of those tasks that they predict workers to sort into. In the generalized Roy model, also the average non‐pecuniary amenities in each task are identified. I apply this approach to the literature on routine‐biased technical change, a key prediction of which is that task prices should polarize. Empirical results for male workers in U.S. data indicate that abstract and manual tasks' relative prices indeed increased during the 1990s and 2000s.