Abstract: Because of the complexity of process modeling on a hydraulic tube tester, it is difficult to diagnose the process fault in view of its mechanism. On the other hand, the fault diagnosis based on MFDA (multiway Fisher discriminant analysis) has high diagnostic capability because it takes advantage of both normal and fault data in modeling. To solve the problem of uneven-length batches, a stage-based header-even method is proposed according to the characteristics of data sampled from hydrostatic testing process, thus adjusting all the batches to become average length. Then, different MFDA models are developed for every sub-stage that was divided in testing process. Combining with the recursive method, the diagnosis results of three stages are intergrated together to improve the diagnostic precision. The effectiveness of the proposed method is tested and verified by applying it to the fault data of four shuttle valves.