Inspection for machinery industrial products is usually done by a man trusting his sense. Automation of inspection has been desired to overcome the differences by individuals and the shortage of such skilled workers. This paper describes the characteristic extraction method of four abnormal engine sounds arising from "Large clearance between cam and valve lifter" and others caused by typical engine assembly error, in order to develop engine sound diagnostic system of automotive on mass inspection line. The sound data for both normal and abnormal engine were measured by changing engine revolutions and measuring positions. To extract the features of abnormal engine sound, Linear Prediction Coefficient (LPC) is calculated after suitable filtering at adequate engine revolutions. As a result, we have found the characteristic difference between normal and abnormal engine sound. Detection of abnormal engines has become easier by compressing sound data into LPC to find their characteristic.