The normalized least mean square(NLMS) adaptive filtering method is introduced to get the preprocessing of near-infrared (NIR) spectrum in order to deduct the noise from Near-infrared spectrum. Fifty-one soil samples are served as the target and the application of NLMS adaptive filtering method in NIR spectrum preprocessing is discussed. The result after denoising is related to the real organic content of soil samples, then constructing a model according to this. Experimental results show that the correlation coefficient of the prediction set is improved from 0.8284 to 0.9654, and the root mean square error of prediction (RMSEP) is reduced from 0.3385 to 0.1606 after denoising with NLMS adaptive filter. So NLMS adaptive filter has a good effect in denoising the NIR spectrum. And it is also very useful to make the final model more representative, stable and robust. NLMS adaptive filter provides a new method for near-infrared sprctrum preprocessment.