A harmonic detect method based on BP network in power system is introduced.It resolves the problem learning of Hidden layer and can be used to detect power system harmonic.For the output of the Fourier form of BP network is equal with the Fourier series,the network is designed to be a SISI system.Hidden layer is composed with Fourier trigonometric functions as activation functions.Gradient descent is adopted to adjust weight.To ensure the convergence of the BP network,learning step η is set according to Lyapunov theorem.Simulation result shows that it improves the convergence speed and the precision of calculation.It acquires the harmonic magnitude and phrase precisely and quickly.