General harmonic detecting scheme uses the Fast Fourier Transform (FFT) to detect all the harmonics (such as from 1st harmonic to 50th harmonic). Power consumers often do not care about the specific values of all the harmonics, but only some key harmonics or several overall indicators. For the above reasons, a new harmonic detecting scheme based on the algorithm of BP neural network is presented. It does not need to calculate all the harmonics and can detect individual indicators or overall indicators which are concerned by users. To achieve the above-mentioned detection target, the computation analysis of BP, DFT and FFT algorithm is made, and the superiority of the scheme in terms of computation is proved. After the simulation of the scheme is done using a set of measured harmonic data, the results validate that the above scheme is simple and feasible, and its detecting accuracy is close to that of FFT.