Cloud detection is the first step in studying the role of polar clouds in the global climate system with satellite data. In this paper, the cloud detection algorithm for the Moderate Resolution Imaging Spectrometer (MODIS) is evaluated with model simulations and satellite data collocated with radar/lidar observations at three Arctic and Antarctic stations. Results show that the current MODIS cloud mask algorithm performs well in polar regions during the day but does not detect more than 40% of the cloud cover over the validation sights at night. Two new cloud tests utilizing the 7.2 Am water vapor and 14.2 Am carbon dioxide bands, one new clear-sky test using the 7.2 Am band, and changes to the thresholds of several other tests are described. With the new cloud detection procedure, the misidentification of cloud as clear decreases from 44.2% to 16.3% at the two Arctic stations, and from 19.8% to 2.7% at the Antarctic station.