Abstract At present there is no reliable approach to model and characterize thermal systems using renewable energy for building applications based on experimental data. The results of the existing approaches are valid only for specific conditions (climate type and thermal building properties). The aim of this paper is to present a generic methodology to evaluate the energy performance of such systems. Artificial neural networks (ANNs) have proved to be suitable to tackle such complex problems, particularly when the system to be modelled is compact and cannot be divided up during the testing stage. Reliable “black box” ANN modelling is able to identify global models of the whole system without any advanced knowledge of its internal operating principles. The knowledge of the system’s global inputs and outputs is sufficient. The proposed methodology is applied to evaluate three different Solar Combisystems (SCSs) combined with a gas boiler or a heat pump (HP) as an auxiliary system. The results show that the best ANN models were able to predict with a satisfactory degree of precision, the annual energy consumption of the all systems except the SCS combined with air source HP, in different conditions, based on a learning sequence lasting only 12 days. In fact, the annual energy prediction errors were less than 10% in most cases. The methodology limitations appear in extreme boundary conditions (Barcelona climate) compared to those used during the ANN training process.