For the problem of power quality signal denoising,an adaptive threshold extraction method based on wavelet entropy is presented in this paper. The signal was disposed of wavelet decompositon, the power quality signal at decomposed levels wavelet entropy can be calculated. The wavelet entropy and adaptive threshold are combined to decided the threshold of high frequency coefficients. Compromised improved index threshold function is used to denosied power quality signals, then power quality signals are reconstructed. Compared with denoising the four typical power quality signals(voltage sag signal,voltage transient oscillatory signal, voltage interruption signal and harmonic signal)with the methods of rigrsure threshold,max-min threshold and the proposed method,the simulation results show that the input signal-to-noise ratio is 20dB, for different power quality signal, output signal-to-noise ratio of improved wavelet entropy adaptive threshold is the biggest.