For the prediction of the contents of Cr and Ni in alloy steel samples, multivariate quantitative analysis model was established by optimizing the input variables of support vector machine (SVM) model, which could solve the problem of complex matrix effect of steel alloy samples. The results achieved by the integral intensity of characteristic spectral lines as the different inputs of SVM were found better than the intensity, because integral intensity contains more information of spectral line, spectral width and spectral shape; The multiple characteristic spectral lines of the elements as the inputs of SVM were better than using single element characteristic spectral information, because the influence of matrix effect could be corrected by inputting multivariate spectral information. By combining internal calibration with multivariate calibration, the experiment errors can be reduced and the matrix effect can be calibrated, and the repetition rate and accuracy could be improved. With the introduction of the normalized variable as the support vector machine (SVM) model of input variables, the relative errors of the content prediction of Cr in sample S1 and S2 are 6.58% and 1.12% respectively; and the relative errors of the content prediction of Ni in sample S1 and S2 are 13.4% and 4.71% respectively. The experiment results show that the SVM algorithm can be effectively used for LIBS quantitative analysis by combining internal calibration with multivariate calibration.