This paper compares 10-year backcast projections of U.S. petroleum consumption for the 2000-2009 period based upon a range of functional forms. Although a univariate specification based upon only lagged consumption performed better than most approaches, it provides little value to the policymaker who wants to understand the factors that influence oil consumption and by how much. One structural approach performed considerably better than all other approaches including the univariate model. This specification is the autoregressive distributed lag (ADL) model that allowed oil demand to respond differently (asymmetrically) to price increases and decreases. Certain forms of this equation, however, do well while others perform poorly. Evaluations of backcast projections over different periods and updated frequently can be important byproducts of energy modeling. (C) 2011 Elsevier Ltd. All rights reserved.