Through transforming the fault diagnosis into the pattern classification of fault characteristics, a fault diagnosis method was developed, based on the cloud sorter for liquid-propellant rocket engines. The method synthesized the study on cloud model and the cloud transform, which can deal with the uncertainty and the hard partition of domain boundary for the quantitative attributes which exist in the traditional classification methods. Caculation results with test data show that the method can recognize the fault modes in the startup process of the engine timely and precisely.