This paper presents a novel approach for diagnosing faults in reciprocating compressors by means of vibration analysis. Based on easily available system variables, an identification and transformation method of the relevant channels is introduced to detect damaged valves. The challenge to the discussed method lies in finding an appropriate transformation to map the general information in the data into a solution space to retrieve the information necessary for detecting faults. Finally, the state of the valve is classified by applying pattern recognition techniques to the solution space.