According to the characteristic that benzo (a) pyrene (bap) will fluoresce under the irradiation of light, the ultraviolet fluorescence is used to detect the bap concentration, builds the dectetion of bap concentration experiment system. In order to accurately detect the bap concentration, the data from the experiment need remove noise with the filter contained in the experiment, to get the fluorescent signal effectively. The normalized least mean squared adaptive filter principle is applied to the noise spectral data processing, compared with the wavelet denoising effect, with Matlab to simulate out denoising effect, and analysis of the different denoising effect from different perspectives. After the contrast, using the minimum normalized adaptive filter can keep the non- stationary signal feature, and achieve more ideal denoising effect and better denoising performance. Hence, this method is suitable for the signal processing of fluorescence detection matter concentration.