According to the fault feature of the rotor system, a fault diagnosis of fuzzy entropy and variable predictive mode based on class discriminate (VPMCD) are proposed. The VPMCD approach establishes a prediction model based on the intrinsic relationship between all or part of the extracted characteristic values, and uses this model to recognize the pattern. Firstly, this paper decomposes the rotor vibration signal by EMD to several intrinsic mode functions (IMF). Secondly, it extracts the fuzzy entropy of the first several IMFs, which contain main fault information, to constitute the feature vector matrix. Thirdly, it adopts the VPMCD approach to build the prediction model. Finally, the built VPMCD prediction model is applied to distinguish the working states and fault types of the rotor system. The experimental results prove that the VPMCD approach and fault diagnosis can accurately and effectively identify the work states and fault types of the rotor systems.