Solar surface irradiance evaluations are used by solar energy companies, to calculate the electric power generated by solar plants in the near future and for estimating potential production without reliable measurements. This effort is very important in a free energy market for an efficient management of the solar power production and the electricity grid in order to balance the power demand. A solar irradiance evaluation must be performed by models reproducing all the physical atmospheric processes involved, that is sometimes a more difficult challenge than the wind modeling. In fact the solar generation is strongly dependent on local weather conditions on the whole vertical profile and on the microphysics of water, fundamental for cloud dynamics. At RSE we have considered three different methods for irradiance evaluation, all fed by the same Numerical Weather Prediction model, to perform 3-day ahead forecast for some Italian sites. The first method comes from a modification of the RAMS model in order to produce, among its standard outputs, the three solar surface components (global, diffuse and direct normal irradiation). These components are computed by the original Ritter-Geleyn (RG) scheme provided with the model code. The second method is a Radiative Transfer Model (RTM), developed at RSE, based on the Geleyn-Hollingsworth (GH) scheme and the third is the Kato2 solar solver provided by libRadtran package. The last two methods are fed by the meteorological vertical profiles provided by RAMS model. All the radiative schemes are very sensitive to the atmospheric water content, as cloud cover fraction and liquid/ice water content which contributes to the atmospheric optical depth. The vertical cloud fraction is used by GH in the context of the random overlap criterion, and it can be obtained directly from the meteorological model or evaluated by the same GH. Modeled clouds are permitted only for relative humidity (RH) larger than a tunable critical value RH-dependent. For the Kato2 method, cloud cover is used within the independent pixel approximation, so an integrated columnar value is used without any vertical dependence. We have compared global and diffuse forecasted values, produced by the three methods, with measurements in some Italian sites; in addition a Model Output Statistic (MOS) correction has been applied to reduce errors. The major improvements using MOS are obtained for those configurations characterized by systematic forecast faults, due to either cloud representation or radiative scheme adopted. This work has been financed by the Research Fund for the Italian Electrical System under the Contract Agreement between RSE and the Ministry of Economic Development – General Directorate for Nuclear Energy, Renewable Energy and Energy Efficiency stipulated on July 29, 2009 in compliance with the Decree of March 19, 2009