In order to improve the detection accuracy of soil elements by laser induced breakdown spectroscopy (LIBS), we establish a quantitative analysis model for soil elements of relevance vector machine (RVM). And it is compared with support vector machine (SVM) model and least squares support vector machine (LSSVM) model. The four characteristic lines of Ni element are taken as the analysis lines, after full spectral normalization, RVM, SVM and LSSVM models are established with the training sample set. According to the test results of testing sample sets, we can know that the SVM is inferior to the others model in terms of model prediction accuracy. However, in terms of model stability, LSSVM model is poorer than the others models. Therefore, in the practical applications, the advantages of RVM in model stability and prediction accuracy indicate that it is more suitable for quantitative analysis of laser-induced breakdown spectroscopy.