Using a data set including large‐scale fields and collocated satellite observations (the NEPHOS file), we test the validity of the approach proposed for validating general circulation model type cloud schemes with the future International Satellite Cloud Climatology Project (ISCCP) and First ISCCP Regional Experiment (FIRE) data sets. An ensemble of 2866 atmospheric situations extracted from the NEPHOS file is used to test the European Center for Medium Range Weather Forecasts operational cloud cover scheme and to study the sensitivity to cloud cover of the regional radiation budget over the eastern Atlantic Ocean and the North Sea. This scheme, based on a simple function of the relative humidity, is shown to underestimate the cloud cover (by 26%), especially when stratocumulus clouds are present. This underestimation is related to too small values of the analyzed relative humidity in the model's planetary boundary layer. The vertical gradient of potential temperature is shown to be a better predictor for this low‐level cloudiness. Differences between the observed and generated radiation budget at the top of the atmosphere are induced by the differences in cloud cover. However, differences in the cloud radiative properties introduce some compensating effects, and the mean bias error is only +7 Wm −2. Particularly, the mean value of the shortwave albedo is too high in the model, and its variation as a function of the cloud cover is much too smooth, as the model does not account for the observed increase in the vertical optical depth of clouds when the cloud cover increases.