Previous studies that have tested for a unit root in aggregate energy consumption have potentially reached misleading conclusions because they fail to allow for the possibility that energy consumption might be fractionally integrated and do not distinguish between different types of energy consumption. This study tests for long memory in disaggregated petroleum consumption in the United States using univariate and multivariate Lagrange multiplier (LM) tests for fractional integration. The results point strongly to the need to distinguish between different forms of energy consumption and allow for a generalization of the I(0)/I(1) dichotomy when considering the order of integration of energy consumption. Allowing for short-run dynamics, the univariate test suggests that less than 50% of the series are fractionally integrated. Consistent with expectations the non-stationary series are found to have the highest mean and standard deviation. The multivariate test suggests that petroleum consumption in the commercial and industrial sectors is clearly fractionally integrated when allowing for short-run dynamics, and, as such, exhibits persistent effects, while petroleum consumption in the residential sector is a stationary process. (C) 2009 Elsevier Ltd. All rights reserved.