Graduation date: 2008The operational processing of MODIS imagery to produce the MOD06 cloud product is based on the assumption that cloudy 1-km pixels are overcast. This assumption is examined using a partly cloudy pixel retrieval scheme, which allows for fractional cloud cover within the 1-km pixels. Cloud flags attached to 250-m MODIS imagery data were aggregated to 1-km resolution to compare with fractional cloud cover retrieved using the partly cloudy pixel scheme. For regions containing only single-layered, low-level marine stratocumulus, the fractional cloud cover derived from the 250-m flags was substantially greater than that obtained with the partly cloudy pixel retrievals. When 1-km visible reflectances and 11-μm radiances were interpreted according to the pixel-scale cloud cover fraction, the interpretations obtained with the partly cloudy pixel retrieval scheme were those expected based on radiative transfer theory. Those obtained from the 250-m flags, however, suggested that many of the 1-km pixels that were identified as overcast were only partially cloud covered. These findings were confirmed using 500-m MODIS imagery to identify pixels that were overcast by marine stratocumulus. Based on the spatial uniformity of 1.6- and 2.1-μm reflectances for overcast and cloud-free pixels, reflectances at these wavelengths were used to distinguish between overcast and partly cloudy 500-m pixels. The 250-m cloud flags obtained from the MOD06 cloud product, however, identified many of the partly cloudy pixels as overcast. The overcast assumption made by the MOD06 cloud product also leads to biases in cloud properties such as cloud optical depth and droplet effective radius. Comparisons of cloud layer temperature, optical depth, and droplet effective radius derived from 2.1-μm reflectances show that overcast results agree with what is expected from cloud parcel models. Pixels with partial cloud cover, however, underestimate optical depth and overestimate droplet effective radius and the biases produce trends that counter expectations based on cloud parcel models. These biases become more evident as cloud cover within a pixel decreases.