Aiming at the nonlinear and nonstationary characteristics of the reciprocating compressor fault signal, a method for reciprocating compressor bearing clearance fault feature extraction based on Refined Composite Multiscale Fuzzy Entropy（RCMFE) was proposed. On the basis of refined composite multiscale entropy, combined with the concept of fuzzy entropy, the RCMFE method was proposed, which was used to quantify signal nonlinear characteristics to form fault characteristics. The simulation analysis results of white noise and 1/f noise show that the entropy value of RCMFE is not sensitive to data length, and the probability of undefined entropy is small. The bearing clearance fault of a reciprocating compressor transmission mechanism was the research object. RCMFE was used to realize the feature extraction of fault signal, and compared with multiscale fuzzy entropy and compound multiscale fuzzy entropy. This method has significant feature discrimination and the support vector machine fault identification accuracy rate is higher than the other methods.