The patterns of production underlying the recent rise of global value chains (GVCs) have become increasingly complex. NAFTA supply chains, for example, are now deeply integrated: Using Mexican customs data, I find that exports to the U.S. use a much higher share of American inputs than exports to other countries. However, the conventional framework used to measure GVCs ignores this heterogeneity since it assumes that all output uses the same input mix. I develop a new framework that combines input-output data with additional information on supply chain linkages in order to construct GVCs reflecting the use of inputs observed in the latter. Improving measurement matters quantitatively since it affects both value-added trade measures and counterfactual experiments: I show that incorporating Mexican customs data raises the estimated share of U.S. value in U.S. imported Mexican manufactures from 18% to 30% and amplifies the welfare cost of a NAFTA trade war. Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.