Abstract: Objective In this paper we propose a method for automatic cloud detection,which employs both spec-tral and shape features.This method includes two main steps.First,based on the different spectralfeatures of clouds and snow,an otsu-based method is used for threshold segmentation of a TM Image.Then,clouds are successfully extracted by discriminant analysis using the shape feature.Experimentsindicate that as compared to the traditional threshold-based methods,the proposed method can effec-tively detect clouds with an improved automatization level.