For the low accuracy of the existing power system harmonic signal detection methods, this paper proposes a novel algorithm based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and Teager energy operator algorithm. Firstly, the CEEMDAN is utilized to decompose harmonic signal, to get the intrinsic mode function (IMF) components which have different local characteristic time scales. Secondly, to obtain the true harmonic components of IMF, EEMD denoising method and correlation criterion method are improved to remove respectively the noise signal and the false component. Lastly, instantaneous amplitudes and frequencies of the true harmonic components are calculated by the Teager energy operator, which can obtain harmonic signal energy spectrum information accurately. The method makes full use of the local adaptability of CEEMDAN and the rapid response characteristics of Teager energy operator. EMD, EEMD and CEEMDAN are combined respectively with the Teager energy operator to detect harmonic, and by contrasting to the detection results, not only the method has higher detection accuracy, but also the feasibility and validity of the proposed method is verified.