Abstract: Pipeline leak identification and detection is usually interfered by the noise contained in acoustic signals of natural gas pipelines. In this paper, therefore, a novel noise reduction method called Small Approximations Removal by Means (SARM) was proposed to remove the noise from acoustic signals. In the SARM method, the reconstruction coefficient of the low-frequency wavelets is taken as the mean line, so the good time-frequency resolution performance of wavelet analysis (WA) is inherited. And the small fluctuations around the mean line are removed. In this way, the curve is very smooth while the mutation characteristics are well preserved, indicating good noise reduction effect. Based on the case study, SARM is superior to Wavelet Packet (WP), WA and Singular Value Decomposition (SVD) in terms of noise reduction effect. It provides technical support for leak detection and location of natural gas pipelines.