The impact of modifying two physical parametrizations in the ECHAM5 atmospheric general circulation model (GCM) is reported. First, a diagnostic (relative humidity-based) cloud fraction scheme is replaced by one based on a prognostic description of the subgrid-scale distribution of total water content (the Tompkins scheme). Second, the subgrid-scale information provided by the Tompkins scheme is introduced into radiation calculations using the Monte Carlo Independent Column Approximation (McICA). Experiments are carried out in three model configurations: (1) ECHAM5 with prescribed distributions of sea-surface temperature and sea ice; (2) ECHAM5 coupled to a mixed-layer ocean model; and (3) ECHAM5 coupled to the MPIOM ocean GCM. The primary direct impact of replacing the RH-based cloud fraction scheme by the Tompkins scheme is an increase in very low cloudiness, mainly at mid and high latitudes, along with a reduction in mid-level cloudiness. The most notable effect of using McICA is a strengthening of the negative short-wave cloud radiative effect, without substantial effects on cloudiness. However, when compared to observational data, all model versions perform in essence equally well. For all of them, cloud field statistical properties show substantial differences from the International Satellite Cloud Climatology Project data; in particular, there is a general lack of low- and mid-level clouds with low optical depth. The differences in temperature, precipitation and sea-level pressure between the model versions are rather small. However, in spite of similar performance for present climate, the different model versions show marked differences in their response to increased atmospheric CO2. Copyright © 2010 Royal Meteorological Society