Accurate description of the cloud vertical layer structure is of primary importance for the Numerical Weather Prediction Model to get accurate parameters like longwave, shortwave radiation fluxes, radiation heating rate profile, cloud reflection, cloud radiation effect. However, the cloud vertical structure can not be validated by the data observed on the ground. The occurrence frequencies of high, middle, low clouds and cloud top pressure retrieved from a uptodate algorithm (classifying cloud vertical layer structure), applied with MODIS cloud product MOD06 (Collection 5), are used to evaluate the results from the NCEP North American Mesoscale model (NAM). The evaluation results from the scope of (153° to 48°W, 12° to 62°N), July to October, 2006 show: (1) The distribution patterns of high, middle, and low clouds retrieved from satellite products and model are similar, especially for high clouds. Over the tropical Pacific, the model shows greater high cloud amount than the satellite scene. The model data shows much more low cloud cover over Mexico, the North American continent, and the Atlantic Ocean, and the middle cloud cover is less than high cloud cover. (2) There is the least difference in the occurrence frequencies of middle clouds between satellite retrieval and model results. Peaks for high and low clouds seen in NAM outputs are sharper and higher than those from the satellite retrieval. (3) The latitudemean middle and low cloud amounts from the model are greater than those generated form the satellite retrieval and more noticeably for low clouds. The cloud parameterization in NCEP NAM needs to be improved.