This paper proposes a novel de-noising algorithm based on ensemble empirical mode decomposition (EEMD) and the variable step size least mean square (VS-LMS) adaptive filter. The noise of the high frequency part of spectrum will be removed through EEMD, and then the VS-LMS algorithm is utilized for overall de-noising. The EEMD combined with VS-LMS algorithm can not only preserve the detail and envelope of the effective signal, but also improve the system stability. When the method is used on pure R6G, the signal-to-noise ratio (SNR) of Raman spectrum is lower than 10 dB. The de-noising superiority of the proposed method in Raman spectrum can be verified by three evaluation standards of SNR, root mean square error (RMSE) and the correlation coefficient ρ.