Ensemble empirical mode decomposition (EEMD) can restrain the mode mixing of EMD in a certain extent, however, the trade-offs are increasing the calculation and losing the completeness for the white noise unneutralized completely. Complementary EEMD (CEEMD) adds the white noises in pairs, which can obtain the same results as the EEMD, but the white noises added are neutralized completely. Based on the research of CEEMD and permutation entropy for randomicity detecting, a modified EEMD (MEEMD) is present in this paper. MEEMD decomposes the data using the EMD after the abnormal components are separated using ensemble and average method, and not only can restrain the mode mixing effectively, but also can narrow the reconstruction error and decrease the computation amount. By analyzing the simulation and real signals, the results indicate that MEEMD is feasible and effective in restraining mode mixing of EMD.