The study presents a comparison and validation of 3 state of the art-global-to-beam irradiance conversion models against ground measurements from 22 sites covering a wide range of latitudes, altitudes and climates. One of the 3 models takes into account a climatic turbidity and gives slightly better results in terms of bias and precision. On a hourly time step basis, the validation on over 100,000 station-hour values shows that the normal beam irradiance can be evaluated from the global horizontal component with a negligible bias and a precision of 85 W m(-2) or 23%. When the models are used on data with a shorter time step, the performance decreases, but remains acceptable. With a modification of the DirIndex (or the DirInt) model, the performance can slightly be increased and reach a bias of 2% and a precision of 28% for 10 min data. (c) 2007 Elsevier Ltd. All rights reserved.