Abstract: A forward-modeling about the spectrum characteristics in multispectral wavebands of principal types of clouds has been conducted, and their distributions in different two-dimensional (2-D) spectral space are analyzed. Satellite data in multispectral wavebands from Multi-function Transport Satellites (MTSAT-2) are compared with cloud classification products from CloudSat Cloud-Profiling Radar (CPR) to produce data pairs containing the spectrum characteristics of principal types of cloud. On the basis of above work, a method named Unit-Feature Spatial Classification Method (UFSM) is introduced to retrieve the distribution range and location of high-frequency feature points of clouds in the 2-D spectral space. Two different types of statistical methods, i.e. Minimum Distance Classification and Maximum Likelihood Estimation, are then applied to the results from forward-modeling and inversion in the cloud classification experiment. The UFSM is finally adjusted to bring out comparatively accurate cloud-classification criterions that can be used over the entire day. Results indicate it was feasible to distinguish high, middle, and low clouds, and clouds with/without vertical development according to the spectral characteristics. As a result, clouds dissipation and movement can be monitored reliably and continuously.