Abstract: The content of soil organic matter (SOM) could be obtained rapidly and accurately by using hyperspectral technology. But the model accuracy and stability are different by using different pre-processing transformations and multivariate techniques. In order to identify the best model to predict SOM, in this experiment, the spectral data of paddy soil in Huangchuan County of Henan Province was measured using ASD FieldSpec 3 Hi- Res hyperspectral meter, two multivariate techniques (Multiple Stepwise linear Regression (MSLR), Partial Least Squares Regression (PLSR)) and eighteen pre-processing transformations of spectra data were compared with the aim of identifying the predictive effects of SOM. The results showed that the SavitzkyGolay first derivative using binomial polynomial and three smooth points (SGF3- 2) was the best preprocessing transformation method，either using the MSLR model or PLSR model the predictive model had both low errors and high accuracy. Compared to using MSLR method, the use of PLSR method can obtain more robust prediction model. The combination of PLSR multivariate technique and SGF3- 2 pre- processing transformation provided the best prediction model with the root mean square error for the validation set RMSEv=0.036 and coefficient determination for the validation set Rv² =0.89. To choose the suit multivariate technique combined with pre-processing transformation can improve model accuracy. These different methods used in this study can also be applied to similar soil models for selection.