Taking detection of soil organic matter content as an example, this paper discussed the application method of empirical mode decomposition (EMD) for processing soil near-infrared (NIR) differential spectrum using empirical mode decomposition. And the principles and steps of processing were proposed. Then the soil organic matter content was calculated based on the de-noised spectrum, and it was compared with the result from the nine-point smoothing method and wavelet method. Experimental results showed that the SNR was improved from 3dB to 10dB, and the root mean square error of between raw signal and de-noised signal were reduced to 0.901 from 2.972. The correlation ratio of the prediction set was improved to 0.9803 from 0.9410, and the RMSEP was reduced to 0.3011 from 0.6702. This improved that EMD is effective to get the pretreatment of NIR spectrum, and the EMD method improves the accuracy of near-infrared spectrum detection of soil organic matter content. 