In this study, several operational parameters of a solar energy system are predicted through using a multistage ANN model. To achieve the best design of this model, three different back-propagation learning algorithms, i.e. Levenberg-Marquardt (LM), Pola-Riber Conjugate Gradient (CGP) and the Scaled Conjugate Gradient (SCG) are utilized. Further, to validate the ANN results, some experimental tests have been done in winter 2016 on a solar space heating system (SSHS) equipped with a parabolic trough collector (PTC). In the proposed model, ANN comprises three consecutive stages, while the outputs of each one are considered to be the inputs of the next. Results show that the maximum error rate in Stages 1, 2, and 3 has occurred in the LM algorithm with respectively 10, 6, and 10 neurons. Moreover, the best obtained determination coefficient of all stages belongs to the total system efficiency and has the value 0.999934 for LM-10. As a result, the multistage ANN model can simply forecast operational parameters of the solar energy systems with high accuracy. Manuscript profile