According to the characteristics of X-ray fluorescence spectrum and visible near infrared spectrum for soil in Nanji area of Poyang Lake,three quantitative analysis models for data fusion,including equal right fusion,co-addition fusion and outer product fusion based on least squares vector machine(LS-SVM) were established.Results showed that the models for equal right fusion and outer product fusion have better accuracy and stability than the single spectral quantitative analysis model has,in which the model for outer product fusion exhibits the best performance with a determination coefficient(R2) of 0.85,a root mean squared error(RMSEC) of 0.09,a root mean square error of prediction(RMSEP) of 0.06 and a relative percent deviation(RPD) of 2.41,satisfying the detection requirements.With the advantages of accuracy and reliability,the developed method could provide a reference for the study of soil heavy metal classification and grading method in China.