Integrated assessment models (IAMs) are increasingly used to evaluate carbon policy impacts on energy structure, but different models can yield considerably different results. This paper seeks to frame model results for policymakers and other consumers of model outputs. In this analysis we compare three models: GCAM, MERGE, and EPPA. We apply diagnostic carbon price scenarios, such as ramps and shocks, to identify key differences in model behavior that drive inter-model variability in projected policy impacts on the U.S. energy system. We report model results using several economic parameterizations and find that variation in carbon emissions across models results primarily from differences in carbon intensity of energy supply. These differences arise because models include different low-carbon energy technology options and vary widely in how flexible the electricity supply sector is at adapting to a change in policy. The timing of emissions abatement is also strongly influenced by whether the model is a simulation or an inter-temporal optimization scheme and the amount of foresight exhibited in the model. Our analysis demonstrates the usefulness of novel IAM diagnostic indicators and clarifies model features that are highly relevant for consumers of model results. (C) 2014 Elsevier Ltd. All rights reserved.