. This paper proposes an approach of fault feature extraction for reciprocating compressor gas valves based on theory of cyclic statistics. First, the strength and weakness of the third-order cyclic statistics in extracting signal features are investigated by simulation signals. Since vibration signals for reciprocating compressor gas valves are of typical cyclic stationary, a new method of fault feature extraction is then proposed based on the simulation results. The method utilizes the cyclic bi-spectrum to extract fault features for the corresponding frequencies. The results show that the cyclic bi-spectrum characteristics for typical faults of gas valves are apparently different, and that the typical faults of reciprocating compressor gas valves can be diagnosed exactly. So the new method proposed in this paper is effective and feasible.