A method was developed for automated classification of surface and cloud types using Moderate-Resolution Imaging Spectroradiometer (MODIS) radiance measurements. The MODIS cloud mask is used to define the training sets. Surface and cloud type classification is based on the maximum likelihood (ML) classification method. Classified results then define training sets for another iteration. Iterations end when the number of pixels switching classes becomes smaller than a predetermined number or when other criteria are met. The final class mean gravity values in the spectral domain are used for class identification and a final 1 km resolution classification map is generated for a MODIS granules. This classification procedure refines the cloud mask algorithm, and enables further applications such as clear atmospheric profile or cloud parameter retrievals from MODIS radiance measurements or from the combination of MODIS and other sounder systems such as the Atmospheric Infrared Sounder (AIRS). The advantages of this method are the automated surface and cloud classification independent of radiance or brightness temperature threshold criteria, and interpretation of each class based on the radiative spectral characteristics of different classes. This paper describes the ML classification algorithm and presents daytime MODIS classification and identification results. The classification results are compared with the cloud mask image, visible image, infrared window image and other sources of observations for the initial validation.