Assessing the energy-saving potential in a building stock requires accurate prediction of the energy use in buildings, as well as estimating effects of imposing energy-conservation measures. Bottom-up building physics-based building stock energy models are widely used for this purpose. However, deficient data (e.g. data related to the use of the building) compel modellers to use normative assumptions in its place, thereby compromising the accuracy of building-physics based models. Furthermore, validation of building-physics based building stock energy models is often lacking. In the present study, a hybrid bottom-up building stock energy model was developed in order to overcome the drawbacks of traditional building-physics (engineering) based modelling methods. Using a sample of more than 100,000 residential buildings, individual building-physics based models were calibrated against energy use data in a multiple linear regression setting, thereby providing a novel hybrid bottom-up building stock energy model. Furthermore, embedding building-physics based building energy models in a statistical model made it possible to validate the model by means of common statistical measures. The proposed hybrid model provided significantly more accurate estimates of the energy use in an unseen sample of buildings than a purely building-physics based building stock energy model. Moreover, as the hybrid model included a unique building-physical description of each building in the sample, it could be used for estimating the effect of imposing an arbitrary energy upgrade. This way of setting up a hybrid building stock energy model provides a simple, yet accurate, approach for estimating the energy-saving potential of a building stock that could be used for informing policy makers and other stakeholders. (C) 2019 Elsevier B.V. All rights reserved.