Fault inspection is a key part of ensuring safe operation of freight trains. The development of machine vision technology has resulted in vision-based fault inspection becoming the principal means of fault inspection. An angle cock is an important component in the brake system, and a fault in it could lead to a serious accident. In this paper, we propose an automated vision method to inspect for missing handles on an angle cock during operation of a freight train. Images of the angle cock are acquired and they are analyzed using a proposed gradient encoding histogram and support vector machine that combine to create a detection system. Experimental results show that we achieved a fault detection rate of 99.8% using the proposed system, which represents a good real-time performance and high detection accuracy.