The mosaic approach of Liang and Wang (1997) for the general circulation model (GCM) parameterization of subgrid cloud-radiation interactions is evaluated against the validated cloud-resolving model (CRM) simulation of the Atmospheric Radiation Measurement (ARM) intensive observation period (IOP, June 22--July 17, 1997) at the Southern Great Plains (SGP) site. The CRM-generated cloud statistics determines the required characteristic structure differences between three primary cloud genera (convective, anvil and stratiform). It is demonstrated that the mosaic approach with the CRM cloud statistics well simulates the CRM domain-averaged radiative quantities. The result indicates that the mosaic approach of the cloud overlap based on the cloud genera differing in formation mechanisms and of the optical inhomogeneity by cloud water path scaling can capture, respectively, the dominant effects of the cloud geometric association and optical property variability within a GCM grid.