Experimental investigations on engine test beds represent a significant cost in engine development. To reduce development time and related costs, it is necessary to check the quality of measurements automatically whenever possible directly on the test bed to allow early detection of faults. A fault diagnosis system should provide information about the presence, cause and magnitude of an inconsistency in measurement. The main challenge in developing such a system is to detect the fault quickly and reliably. However, only faults that have actually occurred should be detected because the user will only adopt a system that provides accurate results. This paper presents a methodology for automated fault diagnosis at engine test beds, starting with an explanation of the general procedure. Next, the methods applied for fault detection are introduced. The theoretical principles underlying these physical and statistical methods are described and their functionality is illustrated with clear examples. Using an example with simulated faults, it is shown that the best results can be obtained with a combination of physical and statistical methods. Finally, results from real test bed operation are provided to demonstrate the workings and practicability of this methodology.