Abstract: Gyro random drift is a remarkable factor that can affect the precision of INS(Inertial Navigation System). Wavelet de-noising method is poor in coping with exceptional noise, and it depends greatly on the selection of wavelet base and decomposition scale. Empirical Mode Decomposition (EMD) de-noising method for gyro signal is presented. The signal is decomposed into an Intrinsic Mode Function (IMF) group. Based on this group, IMFs of exceptional noise are first disposed by 2sigma criterion and then the number of IMFs of high frequency noise is determined by correlation coefficient. The de-noising process is finally done by removing the noisy IMFs. Detailed comparison between EMD method and wavelet method is given. Overlapping Allan variance is used to analyze the effect of them, and the applicable ability of EMD method is tested through an INS calculation. It is shown that EMD method outperforms wavelet method in removing exceptional noise and is more efficient in weakening random drift.