A novel approach is proposed for cloud detection in satellite cloud image based on the improved Graph Cuts Model.10 gray features and 80 Gabor texture features were selected from two channels of FY-2C satellite image(infrared channel 1 and visible light channel),and the dimensions of feature vector were reduced to 9 by using PCA(principal component analysis).Then the similarity matrix was built up by those feature vectors.By using the spectral graph theoretic framework of Normalized Cuts,the cloud image was divided into 2 parts,the clear sky and the cloudy sky.Compared with the result of surface observation,the average consistency was 86.51%.The results demonstrate that this method is effective in cloud detection.