In the practical condition,the fault signal of gearbox often contain multiple fault information,but the weak fault signal belongs to weak signal compared with strong fault signal and noise,therefore,the extraction of weak fault signal in complex fault is always a difficult point in rotating machinery fault diagnosis.Based on the problems above,considering that MED（ Minimum Entropy Deconvolution） has strong reduction performance of noise,the intrinsic mode function decomposed by VMD（ Variational Mode Decomposition） is distorted in strong noise environment. The VMD decomposition accuracy is determined by penalty factor α and decomposition times k. A weak fault extraction method based on MED-VMD is proposed. Firstly,the noise reduction of original signal is carried out by MED,further,the initial parameters α and k are set,the noise reduction signal is decomposed through the VMD,the correlation coefficient of the adjacent intrinsic mode function are calculated to determine the best penalty factor α and decomposition times k. Finally,the weak fault information of bearing in gearbox is extracted by the envelope spectrum analysis of the intrinsic modal function.The validity of the proposed method is verified by the simulation signal and the measured data. A new idea for weak fault feature extraction in complex fault in strong noise environment is presented.