The present paper introduces a goal-oriented approach for parameter calibration of thermal building models. In a context of reducing the global energy consumption and greenhouse gas emissions, the goal-oriented method may be used for a robust prediction of a quantity of interest. Contrary to standard inverse methods, we do not aim at identifying all the model parameters in view of simulating the global thermal behavior of the building. Only the model parameters involved in the computation of the quantity of interest are updated. The proposed inverse strategy can lead to low computational time and reduced instrumentation. To validate the method, a first application on a steady state heat transfer problem is presented.