In order to eliminate the error introduced by noise from fixed analyzer method when measuring polarization mode dispersion of optical fiber and improve the accuracy of the polarization mode dispersion measurement system, a novel denoising approach based on wavelet threshold denoising is proposed in this paper. This paper presents the algorithm flow chart based on wavelet threshold denoising and discusses the selection principles of wavelet threshold, wavelet threshold function, mother wavelet and the number of wavelet decomposition layers. We built an experimental platform and compare the measurement results with Fourier transform algorithm and the commercial polarization mode dispersion measurement instrument. Experimental results show that the proposed wavelet threshold denoising method can effectively reduce impact of noise on the measurement results effectively, which is suitable for different types and lengths of test fiber samples. Taking the commercial instrument as the standard reference, the maximum error of the measurement result of this scheme is 2.27%, which improves the accuracy significantly of the polarization mode dispersion measurement results measure by the fixed analyzer.