Aiming at the problem of harmonic detection in power quality analysis, a new method for harmonic detection in power system based on short-time Fourier transform and spectral kurtosis is proposed. As the time-frequency analysis tool, short-time Fourier transform is used to decompose the sampled signal. Meanwhile, the spectrum of standard deviation and spectral kurtosis based on short-time Fourier transform are introduced as auxiliary analysis. The harmonic modal is identified through the spectrum of standard deviation and spectral kurtosis, and the corresponding harmonic components are extracted by the spectrum matrix according to the result of recognition. Furthermore, a method for disturbances location based on singular value decomposition is presented to locate the transient harmonic signals. Results of a simulation verify the efficiency of the proposed method, indicating that it still has high detection accuracy under the condition of low signal noise ratio (SNR), and has better noise immunity and robustness. This work is supported by National Natural Science Foundation of China (No. 51277134).