The detection accuracy of noise signal directly determines the accuracy of grid non-stationary distortion signal measurement. In order to improve the detection accuracy of noise signal, this essay presents a non-stationary distortion noise detection method based on wavelet transform and sine curve fitting. First of all, the grid non-stationary distorted signals are decomposed and reconstructed by wavelet transform to obtain the fundamental signal, then the fundamental signal of the reconstructed is fitted by the Fourier series. The non-stationary distorted signal minus sine signal is the non-stationary distortion noise signal. The method of this paper is validated by the models of the grid impact signal and the grid continuous pulse signal. Results: among the different decomposition level, the correlation coefficient between the obtained fundamental signal, noise signal and the original signal is above 0.95, the correlation coefficient is higher than that of the wavelet transform method, and less affected by different decomposition level. The results show that this method reduces the impact on the detection accuracy by different decomposition level and has higher detection accuracy.