The derivative method can correct baseline effects but also add noise to it. The empirical mode decomposition (EMD) method was proposed to get the pretreatment of near-infrared (NIR) spectrum. Herein,the first derivative NIR spectrum of tobacco was served as the target and the application of EMD in NIR spectrum pretreatment was studied. Experimental results showed that the result after de-noising with EMD was satisfactory. The correlation ratio of the prediction set was improved from 0.9705 to 0.9832,and the RMSE reduced from 0.5606 to 0.3310. And this method obtained a better de-noising effect compared to wavelet transform method. It is concluded that EMD is a useful method to eliminate noise of NIR signals,which makes the final model more representative,stable and robust. EMD provides a new method for near-infrared spectrum pretreatment.