Mode mixing in the Hilbert-Huang transform (HHT) significantly impacts its application. An improved method of endpoint extension after signal preprocessing with morphological filtering is proposed to restrain the mode-mixing. The analysis results demonstrate that the proposed method can extract the intrinsic mode function (IMF) components more quickly and exactly compared with ensemble empirical mode decomposition (EEMD). The test on the real signal analysis of rolling bearing shows that the method can effectively restrain mode-mixing in practical applications.