Abstract: To reduce the impact of noise on infrared spectral signal measurement, a denoising method based on non-subsampled wavelet transform(NSWT) is proposed. In this method, original spectrum signal is decomposed in multi-scale with NSWT. According to the difference between signal and noise in the scales, sub-band coefficients from the decomposition are adjusted by resolving correlated variational partial differential equations. Signal and noise can then be separated in re-composing the sub-bands. Experiments were conducted for denoising performance comparison between traditional wavelet method and our method. The experiment results show that our method is much better in denoising and signal shapes keeping. The mean square error of our method is also less than that of the traditional method.