In order to achieve energy reduction, many buildings implement energy conservation measures (ECMs). Building energy model (BEM) is a useful tool for quantitative analysis of ECMs. However, uncalibrated model will generate misleading results. The BEM has to be calibrated to the as-operated condition based on metered data and building information. In this research, we applied variable selection method to prioritize calibration sequence and strategy. Critical variables that are significant to building energy use are identified through big data learning approaches. In the case study, we applied the methodology to campus office buildings in the U.S.