Abstract: A novel image compression method based on Support Vector Regression（SVR） method is proposed in this paper.On the basis of dividing the image into some sub-blocks，the general Euclid distance is employed to classify these sub-blocks.Then the SVR is employed to establish the model of each class of image sub-block in order to obtain the SVR model’s parameters sets of each class.Thus an image can be described by the sub-block class number，position feature of each block in an image and a group parameters of SVR model.The results of experiment show that this method has high compression ratio，high signal-to-noise ratio and fine resuming effect on subject.Additionally，the compression ratio of this method can be adjusted by changing the dimension of block or threshold.The resuming effect on image can be adjusted by changing the approaching error while establishing the SVR model.