This paper presents a new recognition method for machine parts composed of polygonal and cylindrical faces by using the selective two-stage neural network. An appropriate operation which depends on characteristics of the face is automatically selected and extracts geometric features of the face effeciently in order to recognize the object. The first stage neural network is for identifying object faces and has p-type Fourier Descriptors, area and perimeter as input parameters. The second one is for recognizing the object by a set of aspect graphs which represents the connection among the faces. Consequently, by applying this method to recognize machine parts, the efficiency was improved as compared with ordinary methods.