Hourly models that predict global irradiance on tilted surfaces are validated on one year data acquired in Geneva and Denver. The reflected component is analyzed separately for the site of Geneva. As the beam component is an input parameter for anisotropic transposition models, three ways are investigated: the transposition based on measured beam irradiance, the model chain based on the only global irradiance input data, and the transposition of synthetic hourly data generated from monthly values. The main conclusions confirm the results published in the literature: - if the reflected component is known, the tilted irradiance can be evaluated with a low bias of some percents, and a precision around 11% for the best model, - the bias is not influenced by the use of a diffuse fraction model to obtain the beam component from the measured global, while the dispersion is higher and reaches 22% for the best model, - the knowledge of the reflected irradiance has a great importance on the overall bias when dealing with high tilt angles, its error is directly reported on the model bias.