In order to promote the effect of empirical mode decomposition (EMD) for the harmonic detection, this paper uses the ensemble empirical mode decomposition (EEMD) to eliminate the modal mixing of EMD harmonic detection. After studying, it is found that the noise in the sampling signal has great influence on the EEMD harmonic decomposition. This paper proposes an EEMD harmonic detection method based on a new wavelet threshold denoising pretreatment. This method adopts transforming wavelet coefficients to select threshold accurately, then uses the combination of hard and soft threshold and finally eliminates random noise. After simulation and analysis, this method can effectively eliminate the influence of random noise on harmonic detection and improve the accuracy and applicability of EEMD harmonic detection, and meanwhile, compared with the original EEMD algorithm, the decomposition rate increases about 3.8 times and the correlation between the effective components and the original signals increases by 22.5% .