Soil moisture has strong absorption in near infrared spectroscopy (NIRS) and causes interference in the prediction of the soil organic matter (SOM) content. In this paper, 41 dry soil samples were used to establish the SOM calibration model by PLSR, and 9 samples were used as the prediction set. All soil samples were rewetted to four different moisture contents (5%, 10%, 15% and 17%). The slope/bias (S/B) and direct standardization (DS) algorithms were used to correct SOM prediction results and whole-spectra obtained by different moisture content, eliminating the differences caused by soil moisture. The results showed that the bias reduced and prediction performances of the model were improved, with Rp higher than 0.89 and RMSEP lower than 0.885%. The study indicated that S/B and DS algorithm corrections could effectively remove the influence of soil moisture in NIRS and improve the accuracy of SOM predictions.