Clouds exert a strong influence on the distribution of heating and cooling at the Earth's surface and within the atmosphere. This has long been recognized in climate studies and in the development of General Circulation Models'2 (0CM), much less in Numerical Weather Prediction (NWP), until the extension of useful forecast length and the diversification in the products range has given new impulse to the development of more accurate parametrizations of diabatic processes3. When the new ECMWF prognostic cloud scheme becomes operational, the ECMWF model will acquire some capability to assimilate cloud-related information. To prepare further development, calibrated and earth located, or raw, Tiros-N Operational Yertical Sounder (TOYS) radiances are used as a diagnostic tool to evaluate the ability of sequences of short range forecasts, from different models, to simulate some of the features in the measured data. The present exercise is therefore quite different from the diagnostics which employ long term integrations of a model to test its average properties against some independent data set. The main difference between the raw TOYS radiances used in the present exercise and the radiances used operationally at ECMWF is the cloud clearing process, whose aim is to identify and eliminate the data affected by clouds. Other differences will be discussed in section 2. Since the signature of clouds on upwelling atmospheric radiance is quite marked at visible and infrared wavelengths, inadequacies in model representations of the three dimensional structure of cloud cover and cloud liquid water are easily identified. One can therefore make a detailed examination of aspects of the hydrological cycle and energetics in current NWP models and in GCMs. Although the importance of clouds in NWP lies mainly in their radiative effects, it is common practice to compare model cloud properties, such as cloud amount (either total or for thick layers) with "equivalent" quantities retrieved from data sets which are independent of the model, and are frequently derived from satellite radiance data68. In such a comparison the same name (for example total cloud amount) may be given to quite distinct quantities, obtained using different underlying assumptions9. Great care is needed when comparing these satellite products among themselves and with model products'°. On the other hand when simulated radiances, obtained using all pertinent model information, are compared to measurements, one needs to be aware of the assumptions used in the simulations themselves. The evaluation of the errors is in both cases a delicate procedure, but the error estimation is particularly difficult for retrieved products such as total cloud cover. In the following the results of a comparison between a selected set of raw TOYS radiances and simulations using the required model output fields is presented.