Three kinds of soil from Hubei province was studied, PLS and cross calibration method was employed to establish soil moisture analysis model. The result indicates that the model’s decision coefficient was 0.995，and RMSECV was 0.801%, the model’s forecast decision coefficient was 0.992, the RMSEP was 0.912%. Using the model to forecast loessial soil on Loess Plateau region, the error was greater than 4%. At last, the study of the Hubei yellow brown earth, paddy field soil and alluvial soil samples collected from Fourier transform parameters and characteristics of soil moisture to establish PLS prediction model, model decision coefficient was 0.988, and RMSECV is 1.106%. And the error of model prediction of loessial soil was less than 2%, the accuracy improved more greatly than traditional model.