According to the strong nonstationarity, nonlinearity,and multicomponent coupling characteristics of reciprocating compressor vibration signal,a feature extraction method based on compound interpolation local mean decomposition method (CIELMD) and refined composite multiscale fuzzy entropy(RCMFE)was proposed. The CIELMD method was used to decompose the fault signals of different bearing clearances, and the highlighted PF components which contained the main information of fault state were chosen with the correlation coefficient.The PF components were quantitatively described by the RCMFE method to form the state characteristic matrix. Furthermore, to solve the problem of information redundancy,the memetic algorithm was further used to optimize the elements with the largest average sample distance in the matrix to form a feature vector with striking separability. Finally, the experimental results of the reciprocating compressor bearing clearance fault signal shows that the method has strong separability in extracting fault features and high fault identification accuracy.