In order to make better use of laser induced breakdown spectroscopy (LIBS) in the liquid steel composition monitoring, a model of support vector machine based on radial basis function (RBF) kernel function is established by using particle swarm optimized support vector machine. In order to reduce the interference of instrument and environment, integral intensity of spectral line and Fe normalization is used. The experimental data is subjected to principal component analysis to carry out quantitative analysis of the concentration of Mn element in molten steel, it is obtained that the mean square error (MSE) is 0.599% , the relative standard deviation (RSD) is 8.26%, the correlation coefficient is 0.997. The results show that the particle swarm optimized support vector machine regression method can be used to analyze LIBS of liquid steel composition, its analytical performance is improved compared with traditional calibration methods.