A detailed understanding of material flows is needed to target increased material efficiency and circular economy. In this article, the U.S. steel flow is modeled as a series of nodes representing processes and products. An easily updatable nonlinear least squares optimization is used to reconcile the inconsistencies across 293 collated data records on flows through and between the nodes. The data come from an integrated analysis that includes top-down estimates of steel flow from trade bodies and government statistical agencies, bottom-up estimates of the steel embedded in products based on production statistics and bills of materials, and the mass of imports and exports based on international money flow. A weighting methodology is used to consistently assign confidence scores to the data, and the optimization is used to achieve mass balance and minimize the sum of the squares of the weighted residuals. The results indicate that yield improvement efforts should focus on sheet metal forming in the car industry, which accounts for nearly half of all generated fabrication scrap. The quantity of end-of-life scrap exported and land-filled is greater than the quantity of steel products imported. Increased domestic recycling of end-of-life scrap might displace around a third of these imports.