One of the possible ways to check the adequacy of the physical description of meteorological elements in global climate models (GCMs) is to compare the statistical structure of these elements reproduced by models with empirical data from the world climate observational system. The success in GCM development warranted a further step in this assessment. The description of the meteorological element in the model can be considered adequate if, with a proper reproduction of the mean and variability of this element (as shown by the observational system), the model properly reproduces the internal relationships between this element and other climatic variables (as observed during the past several decades). Therefore, to distinguish more reliable models, the authors suggest first analyzing these relationships, ‘‘the behavior of the climatic system,’’ using observational data and then testing the GCMs’ output against this behavior. In this paper, the authors calculated a set of statistics from synoptic data of the past several decades and compared them with the outputs of seven GCMs participating in the Atmospheric Model Intercomparison Project (AMIP), focusing on cloud cover, one of the major trouble spots for which parameterizations are still not well established, and its interaction with other meteorological fields. Differences between long-term mean values of surface air temperature and atmospheric humidity for average and clear sky or for average and overcast conditions characterize the long-term noncausal associations between these two elements and total cloud cover. Not all the GCMs reproduce these associations properly. For example, there was a general agreement in reproducing mean daily cloud–temperature associations in the cold season among all models tested, but large discrepancies between empirical data and some models are found for summer conditions. A correct reproduction of the diurnal cycle of cloud–temperature associations in the warm season is still a major challenge for two of the GCMs that were tested.