In this study, a new theoretical approach offering a good prediction of the performance of photovoltaic modules/strings/arrays was developed. The approach implemented in the Matlab/Simulink environment, is based on Levenberg Marquardt (LM) algorithm and was used to estimate the parameters of five electrical models selected among the most used ones. First the problem of initial guess is considered. For parameters initial values, an analysis of six was performed and lead to the choice of the group which offers the best trade-off between accuracy and speed of calculation. To validate the effectiveness of the proposed approach, the five-parameter model (L5P) is used for a comparison with both a deterministic method and two heuristics methods. The results clearly show that, the accuracy achieved with LM method is comparable to the deterministic one, but higher than that of the heuristics methods. Furthermore, the five selected electrical models have been evaluated on four different PV modules technologies. The I-V characteristic curves obtained under Standard Test Conditions (STC) by each of them, are compared to the manufacturer data. It was shown that, when the LM algorithm is used, the fives electrical models predict the behavior of the photovoltaic silicon modules with close accuracy. The best trade-off is achieved with L5P model. This result is confirmed by the theoretical estimation of solar energy production performed for three real power plants by using the fives models. The maximum difference between calculated and measured energy is around 14%. (C) 2014 Elsevier Ltd. All rights reserved.