Axle box bearing is a key component of rail vehicles. Its health condition greatly affects operation safety of the entire system of rail vehicles. Understanding fault behaviors is greatly related to successful development of fault diagnosis methods. However, most previous studies focus on artificial faults, and real faults that were generated naturally from field operation have rarely been investigated. Considering this challenge, five axle box bearings with real faults have been collected from field operation rail vehicles. Their vibration response under various working conditions has been collected. Based on the vibration data, fault behaviors are investigated from a viewpoint of classification by using the C4.5 decision tree. The results about influence of speed, load, and fault mode to fault diagnosis make a better understanding about real bearing faults for rail vehicles.