The accuracy of hyperspectral multivariate information extraction models for soil is directly related to the role of spectroscopy in the future.The paper has summarized the current methods which are suitable for soil spectral processing,including abnormal screening,smooth denoising,resampling,transform and quantitative methods on the basis of the lists of soil samples preparation, physical and chemical components determination and soil spectral data acquisition.In this paper,8 modeling methods such as partial least squares,principal component analysis and multiple stepwise regression are compared.The paper draws up a variety of spectral information extraction models which have achieved good results according to the organic matter,moisture content,salinity,heavy metals and other components.The information indicated by spectrum can provide a quick indication information for soil composition,and the information extraction model established based on the measured data is the theoretical basis for software development,instrument development and soil quality evaluation.