Numerous problems can occur for an investigator of larger datasets, e.g. how to handle dimensionality, many variables and few observations, few variables and many observations, correlations, missing data, noise and to extract information from all data simultaneously. Multivariate analysis (MVA) is an established method for dealing with such problems. In this work, we introduce a methodology based on MVA, which was developed to model the building energy performance from the perspective of the property holder. Data from a Swedish database of 500 buildings, which recently has been compiled and is under expansion, was used for the investigation. The available data consists of building specific information and consumption data, monitored on a monthly basis, reported by the property holder. Electrical consumption for lighting and appliances is paid by the tenants in Sweden, and is thus lacking in the database. This means that the data base just include the part of the total energy use that is paid by the property holder. With the overall goal to assess the energy use paid by the property holders, a methodology is suggested for estimating the electrical energy paid by the tenants. At this early stage of our work, we found that the used methodology gives a fairly robust model and that the interpretation of the model is believed to be accurate in terms of comparing the energy use between different buildings.