In the present paper, a wavelet modulus maximum shift-related filter algorithm is proposed for denoising the absorption spectra. Firstly, using the wavelet transform modulus maxima theory, useful signal components and noise components of the binary wavelet coefficients of absorption spectra are identified. Then, the useful signal components are aligned in the wave number domain to correct the “drifting” of modulus maximum across the scales, and the noise components are smoothed. Finally, according to the wavelet interscale dependencies obtained by multiplying the adjacent wavelet subbands preprocessed by the above procedure, important features of signal are enhanced while noises are attenuated further. Compared with the traditional spatially selective noise filtration technique proposed by Xu et al and adaptive multiscale products thresholding technique proposed by Paul Bao et al, the proposed wavelet modulus maximum shift-related filter algorithm has several advantages. First, it does not need estimate of the noise intensity, which could avoid the error introducing and the complex calculation. Meanwhile, there is not iterative calculation in the proposed algorithm, which could eliminate the risk of slow convergence or no convergence of the algorithm. Furthermore, the “drifting” of modulus maximum across the scales could be corrected in the proposed algorithm, which could make up for the loss of the spectrum information caused by the “drifting” phenomenon. Experiments show that the proposed scheme can effectively suppress noise and preserve the useful components in the infrared absorption spectra of SF6 gas.