The vertical distribution of clouds significantly affects the radiative heating/cooling distributions. Consequently, the general circulation model (GCM) simulated climate can be sensitive to the treatment of cloud overlap, which is related to the subgrid scale cloud vertical geometric association. Here we developed and tested a mosaic treatment that can explicitly consider the cloud association in the GCM radiation parameterization. The effects of this treatment on radiative forcing and climate simulations were studied by comparing the results with those using the random overlap, which assumes clouds are independent and therefore yields a larger effective cloudiness. The mosaic treatment calculates a significantly different atmospheric radiative heating/cooling distribution caused mainly by the changes in infrared radiation. In the tropics it yields a heating in the upper troposphere and a cooling in the lower troposphere especially near the surface; opposite changes are calculated in the middle to high latitudes. Because of a smaller effective cloudiness the mosaic treatment calculates less infrared downward radiation reaching the surface, which is partially compensated by increased incident solar radiation. Differences in the climate responses are substantial, with several major model biases corrected by the mosaic treatment. For example, the middle to upper troposphere of the tropics and subtropics are warmed by more than 3 ¿C throughout the year, and the polar night northern stratosphere becomes much warmer, up to a maximum of 15 ¿C. A more realistic distribution of tropical precipitation is simulated especially over central to east Pacific oceans. Other notable improvements include the decreased convective rainfall and surface evaporation and the increased atmospheric moisture content. The mosaic treatment also enhances significantly the daily variability of surface air temperature and radiation fluxes in the low latitudes.¿ 1997 American Geophysical Union