According to the characters of SPR optical fiber sensor spectrum and the requirement of real-time monitoring, a new noise filtering method, mobile lifting wavelet analysis, is presented in this paper. This method is not based on Fourier transform, but on an algorithm of dividing a complicated noise filtering process into a number of reversible simple processes. It is very fast, correct and does not use additional memory. A model of actual SPR optical fiber sensor spectrum is constructed by superposing a certain intensity of Gaussian white noise on the theoretical spectrum, of which the resonant wavelength is 557.70 nm. It applies noise filtering to the simulative spectrum with mobile lifting wavelet analysis based on Haar, CDF(3,1), DD(4,2) and 5/3 mother wavelet respectively and calculates the resonant wavelengths again. The results are 556.45, 564.06, 557.27 and 557.91 nm corresponding to each method listed. So a relative error of 0.037 7 percent, obtained after noise filtering with new method based on 5/3 mother wavelet, is the minimum one. It is also lower than 0.430 3 percent obtained after noise filtering with traditional symlet11 wavelet analysis that has been proved to be effective for SPR optical fiber sensor spectrum. At different time gather several spectra of one SPR optical fiber sensor detective system were gathered and mobile lifting wavelet analysis based on 5/3 mother wavelet was done. The result shows that, the standard deviation of resonant wavelengths is reduced to 1.560 8 from 4.186 7 nm, which is calculated before noise filtering. As expected, this result is better than doing the same experiment with traditional symlet11 wavelet analysis, which only reduces the standard deviation to 2.725 3 from 4.186 7 nm. The research shows that mobile lifting wavelet analysis significantly suppresses the system noises, reduces noise influence on the gathering of resonant wavelength information from SPR optical fiber sensor spectrum and gives a guarantee to actual accurate detection with SPR optical fiber sensor.