Abstract: The vibration signals of the vacuum pump have non-stationary and non-linear characteristics, mixed with a large number of ambient noise signals. It is difficult to extract and analyze the characteristic signals directly, as a challenge for the online fault diagnosing of the vacuum pump. To deal with that problem, a wavelet threshold de-noising method is proposed based on the ensemble empirical mode decomposition (EEMD) for the vacuum pump vibration signals. Firstly, the vibration signal is decomposed into several intrinsic modal functions (IMF), with a residual by the EEMD. Then, the normalized autocorrelation function is introduced to screen the IMF components, and the filtered IMF components are subjected to the wavelet threshold de-noising. Finally, the de-noised IMF components are reconstructed with the unprocessed IMF components and the residual to obtain the de-noised vibration signal of the vacuum pump. The noise reduction result of the simulation signals and the experimental signals shows that the proposed method is superior to the existing methods with a raised SNR and a diminished RMSE, which provides a new way for de-noising the vacuum pump vibration signals.