Accurate building performance assessment is necessary for the design of efficient energy retrofit operations and to foster the development of energy performance contracts. An important barrier however is that simulation tools fail to accurately predict the actual energy consumption. We present a methodology combining thermal sensor output and inverse algorithms to determine the key parameters of a multizone thermal model. The method yields calibrated thermal models that are among the most detailed ones in the literature dealing with building thermal identification. We evaluate the accuracy of the resulting thermal model through the computation of the energy consumption and the reconstruction of the main energy flux. Our method enables one to reduce standard uncertainties in the thermal state and in the quantities of interest by more than 1 order of magnitude.