Subtracted blood volume spectrometry was employed to the noninvasive biochemical examination with near infrared spectroscopy(NIRS) to eliminate the influence of tissues.To raise the Signal to Noise Ratio(SNR) of the NIRS pulse wave signal needed by the spectral subtraction, an effective adaptive filter method was proposed to process the pulse wave signal. The principles of Least Mean Square(LMS) adaptive filter were described, and a new adaptive filtering way fit for the pulse wave signal of this experiment was proposed. Then, a 16-pixel near-infrared pulse wave acquisition system made by ourselves was used to collect the pulse wave signals of human body. Finally, the proposed adaptive filtering way was used to process the NIRS pulse wave signals and analyze the results. The result shows that the noise level of blood spectrum has reduced from 800 μAU to 12 μAU after spectral subtraction by using the proposed method, and the related coefficient of pulse wave of adjacent wavelength has raised from 0.994 0 to 0.999 9. The analysed result verifies that the method is effective in the NIR noninvasive biochemical examination area.