The inversion problem of MFL detection is mainly to reverse the contour size of defects through MFL detection signals. The inverse problem of MFL detection is ill posed, so the accuracy of MFL detection signals has a great impact on the accuracy of inversion results. There are a lot of complex noise signals in the actual detection of MFL signals, which can not be directly used for inversion calculation. Based on the median filter algorithm, this paper proposes a multilevel filter algorithm for MFL signal processing, which can eliminate the possible high frequency noise in MFL signal through the improved median filter algorithm of level 2, and then carry out the median filter of level 2 to eliminate the impact of low frequency baseline drift. The experimental platform is built to collect MFL signals and the filtering algorithm is studied, which proves that the filtering algorithm has good noise reduction effect on the original MFL signals.