In the clustering analysis of test data of liquid rocket engine, in order to solve the problem that there is insignificant difference between fault sample and normal sample, the maximum scatter difference criterion was introduced and the maximum scatter difference based clustering algorithm (MSD-CA) was presented. In MSD-CA, the similarity of samples was measured by divergence, and the minimizing of the within-class divergence and maximizing of the between-class divergence were processed together. After that, fuzzy theory was introduced to maximum scatter difference criterion, and the maximum scatter difference based fuzzy clustering algorithm (MSD-FCA) was presented and used to do “soft partition” for test data to improve the precision of cluster. The method is verified with experimental results.