A large amount of noise contained in power quality signals may greatly affect the result of detection in power systems. Based on the cross validation criterion， a new method for suppressing the white noise in power quality signals is proposed. The method can select the optimal threshold value adopting the cost function of Integrated Square Error（ISE） without estimating any parameter of the background noise. Thus it can solve the intrinsic difficulty of selecting the wavelet function in the process of decomposition and the effect of de-noising thereafter is better than that of traditional methods. Simulation results have proved the effectiveness of the method proposed in extracting the abrupt information of disturbing signals and improving the signal-to-noise-ratio（SNR）. Furthermore， the square error（SE） of the de-noised power quality signals is fairly small.