In order to improve measurement accuracy in the polarization mode dispersion (PMD) measurement system by the fixed analyzer method, the noise in the link should be eliminated. Here, a polarization mode dispersion measurement scheme based on empirical mode decomposition (EMD) adaptive filtering to mitigate noise is proposed. According to signal spectral characteristics, it is decomposed into a finite number of intrinsic mode functions (IMFs), which are arranged by frequency from large to small. The consecutive mean square error (CMSE) criterion is used for noise filtering and signal reconstruction, which effectively eliminates the noise impacting on the measurement results. The experiments compare results based on the EMD adaptive filtering method, Wiener filtering method and commercial instruments. The results show that this scheme is suitable for different types and lengths of fibers, and the maximum relative error between the measured result and the reference value is 0.74%. The measurement accuracy is improved obviously.