Combined squared envelope spectrum（CSES） is susceptible to the interference from in-band noise and other band characteristics in high back-ground noise， which leads to the difficulty in extracting the weak fault features of rolling element bearings. A new method based on CSES and minimum entropy deconvolution （MED） is proposed to overcome this problem. Firstly， the proposed method uses spectral kurtosis to select the filtered signals from different frequency bands. Secondly， these selected signals are further filtered using MED for enhancing the fault feature in the frequency band. According to the principle of CSES， the squared envelope spectrum is computed for the signals filtered in previous step. These spectrums are normalized within 0 and 1， then they are combined to obtain an enhanced envelope spectrum with obvious fault characteristics. The simulation and experimental results show that the proposed method can effectively extract the weak fault features of rolling element bearings.