Color scanner characterization method based on support vector machine (SVM) was studied. The feasibility of nonlinear transformation of RGB color space to CIELab color space using SVM regression was analyzed. The results showed that the correlativity is more than 99% between test value and predicted value of L, a, b, among which the average, maximum and minimum color differences are 2.314 3, 5.791 7, and 0.507 3. It was concluded that SVM regression method can satisfy the accuracy requirement of scanner characterization and has the advantages of precision and time for small sample fitting.