The Moderate Resolution Imaging Spectroradiometer (MODIS) on Terra, with its high horizontal resolution and frequent sampling over Arctic and Antarctic regions, provides unique data sets to study clouds and the surface energy balance over snow and ice surfaces. This paper describes a polar cloud mask using MODIS data. The daytime cloud and snow identification methods were developed using theoretical snow bi-directional reflectance models for the MODIS 1.6 and 3.75 micron channels. The model-based polar cloud mask minimizes the need for empirically adjusting the thresholds for a given set of conditions and reduces the error accrued from using single-value thresholds. During night, the MODIS brightness temperature differences (BTD) for 3.75 - 11, 3.75 - 12, 8.55 - 11, and 6.7 - 11 micron are used to detect clouds while snow and ice maps are used to determine snow and ice surfaces. At twilight, the combination of the 1.6 micron reflectance and the 3.75 - 11 micron BTD are used to detect clouds. Examples of the cloud mask results from daytime, nighttime, and twilight data show good agreement with visual interpretation of the imagery. Comparisons of the modeled and observed reflectances for clear snow areas reveal good agreement at 1.6 micron, but 10 - 35% overestimates of the 3.75 micron reflectance by the model. Over the Arctic, the modeled visible reflectance is significantly greater than the observed values. Better agreement is obtained over the Antarctic where snow melt is less significant.