In the present paper, based on the multi-resolution attribute of EEMD (ensemble empirical mode decomposition) method, we presented a new de-noising method for analyzing spectrum, and applied it to process the reflecting spectrum data of 33 soil profiles in the typical oasis located in the middle reaches of the Tarim River. To explore the de-noising effect of EEMD threshold method for reflecting spectrum in soil profiles, we compared EEMD threshold method with wavelet transform method. The results showed that compared with traditional wavelet transform method, the signal to noise ratio (SNR) was improved from 14.836 6 to 34.275 7 dB, and the root mean square error (RMSE) was reduced to 7.240 6×10-6 from 6.786 1×10-5 and the correlation coefficient (r) increased from 0.982 5 to 0.999 8. Therefore, three de-noising effect indicators of EEMD threshold method are better than those of wavelet transform method. This proved that the EEMD threshold method can effectively eliminate the noise of soil-profile spectrum and also preserve the detailed information of the original spectra well. Thus, the analysis precision of the spectrum will be improved. In addition, by contrast with the wavelet threshold method, the EEMD threshold method is adaptive and is fairly reliable. As a new method for spectral pretreatment, the EEMD threshold method will have a good application prospect in spectra de-noising.