© 2016 Elsevier Ltd. Energy assessment of urban buildings has become an active research field due to a large amount of energy consumed in cities as a result of fast urbanization. Hence, it is necessary to determine relative importance of variables for explaining variations of building energy use. However, two commonly used methods (correlation analysis and standardized coefficient) are only suitable for uncorrelated variables. This may not be the case for an extensive urban dataset containing social, economic, and physical variables. Therefore, this study proposes a two-stage approach to handle a large number of correlated variables in urban energy analysis. London has been chosen as a case study to determine influential factors affecting domestic energy use. The first stage applies two fast-computing methods (Genizi measure and correlation-adjusted score) to select important factors. The second stage implements two computationally intensive approaches (Lindeman Merenda Gold and proportional marginal variance decomposition) to further assess relative contributions of explanatory factors selected in the first step from conditional and marginal perspectives. The results indicate that this two-stage approach can deliver reliable results by explicitly accounting for correlations among variables in urban energy assessment.