The effect of vertical resolution on cloudy-sky radiation calculations is investigated through idealized, single-column experiments and extensive tests with a high vertical resolution (100 layers) general circulation model (GCM)-generated data set. As examples of GCM-type radiation codes, the European Center for Medium-Range Weather Forecasts (ECMWF) and Deutscher Wetterdienst (DWD) schemes are considered. The basic assumption in the tests is that vertical discretization distorts the profiles of temperature, absorbing gases, liquid water, and ice only by removing the subgrid-scale details. On the whole, cloudy-sky radiation calculations appear significantly more sensitive to vertical resolution than clear-sky calculations, owing to the following reasons: (1) Both in the longwave and in the shortwave, a critical factor is how the assumed subgrid-scale horizontal distribution of cloud water changes with vertical resolution. This issue also depends on the cloud overlap assumptions, random overlap tending to lead to increasing cloud cover and cloud forcing with improving resolution. (2) Application of maximum-random overlap to effective cloud fractions in the ECMWF longwave scheme leads, in particular, to severe underestimation of cloud forcing at the top of the atmosphere at high resolution. (3) The assumption (in the ECMWF scheme) that cloud layers emit upward (downward) at the exact layer top (bottom) temperature tends to lead to overestimated longwave cloud forcing at coarse resolution; the same also occurs if (4) cloud top (bottom) heights are overestimated (underestimated) at coarse resolution. (5) Cloud optics parameterizations may be dependent on vertical resolution (this affects, to some extent, the shortwave results of both schemes considered). On the positive side, both the longwave and shortwave results of the DWD scheme depend little on vertical resolution if vertical discretization does not distort the cloud properties. ¿ 1999 American Geophysical Union