Estimates of marginal abatement costs for reducing carbon emissions in the United States by the major economic-energy models vary by a factor of five, undermining support for mandatory policies to reduce greenhouse gas emissions. We use meta analysis to explain these cost differences, holding policy regimes constant and focusing on the role of baseline emissions projections and structural characteristics of the models. The results indicate that certain assumptions, like freer trade and greater disaggregation of regions and nonenergy goods, lead to lower estimates of marginal abatement costs, while more disaggregated energy goods raise them. Other choices, like myopic optimization by households or the inclusion of an international finance sector, seem less significant. Nor do emissions baseline differences explain much of the cost differences. Our analysis can help indicate which modeling assumptions are most important to understanding the cost discrepancies and developing consistent modeling practices for policy evaluation.