An efficient cloud classification algorithm is proposed by combing Support Vector Machine (SVM) with fractal geometry, moment feature, discrete cosine transformation and some region features. Fractal dimension describes the complexity and roughness of texture of cloud. The feature vector is constructed by combing fractal dimension, invariant moment, discrete cosine transform descriptors and some shape, texture descriptors. Cloud classification models are built by SVM. The models are used to respectively classify two ones of altocumulus castellanus, altocumulus undulatus and cumulonimbus capillatus.