Aggregate productivity suffers when workers and machines are not matched with their most productive uses. This paper builds a model that features industry-specific markups, industry-specific returns to scale, and establishment-specific distortions and uses it to measure the extent of this misallocation in the economy. Applying the model to restricted U.S. Census microdata on the manufacturing sector suggests that misallocation declined by 13% between 1982 and 2007. The finding of declining misallocation starkly contrasts with the 29% increase implied by the widely used assumptions that all establishments charge the same markup and have constant returns to scale.