How to extract fault features from nonlinear and non-stationary vibration signals of rolling bearings under the influence of complex noise background has become a key issue. Therefore, a diagnosis method based on direct fast iterative filtering (dFIF) was proposed. This technique first makes use of dFIF to adaptively decompose the raw vibration signal into a set of intrinsic mode functions (IMFs). Then for the selection of fault-related modes, a comprehensive criterion referred to as envelope entropy-kurtosis is proposed. Finally, the envelope analysis of the chosen optimal components is used to extract the fault characteristics and judge the fault type. The performance of the proposed method comparing to the most common methods are investigated by the application example study. The result exhibit that the proposed approach can extract the characteristic frequency of the fault vibration signal more rapidly and accurately, and accurately identify the type of the fault.