How to extract weak fault characteristics from vibration signals with noise is a key problem of bearing fault diagnosis. A novel method based on sensitive singular value decomposition（ SSVD） and ensemble empirical mode decomposition（ EEMD） is proposed

aiming at the above problem. SSVD is applied to vibration signals which are analyzed in time domain

SSVD component signals which reflect fault impact characteristics are selected via the sensitive factor

and the corresponding singular values for vibration signal reconstruction are determined by using positioning factor

noise will be removed effectively through above steps. The de-noised signals are decomposed by EEMD

so that intrinsic mode function（ IMF） components which includes sensitive fault information can be selected according to the kurtosis criterion

through which the local weak fault information of vibration signals can be effectively extracted. Instantaneous energy of fault information is calculated by using Teager-Kaiser energy operator（ TKEO）

and the spectrum analysis is performed to obtain fault characteristic frequency

will be used to identify type of fault. The experimental results prove the effectiveness of this method of rolling bearing fault diagnosis.