Scheme evaluation and selection is an optimum selecting and sequencing problem with multi-objective and multi-level. It cant follow single objective function or rule. Meanwhile, these objectives are coupled with each other and the at-tribution information is fuzzy also. It is necessary to find an effective evaluation method which can consider all conditions and restrictions. In this paper, AHP and rough set theory are applied to fuzzy optimization to determine important weight of each attribution. The rough set fuzzy optimum selection is used to eliminate the useless information. Autonomous un-derwater vehicle (AUV) is large-scah systems with many coupled design variables and objective functions. Their scheme evaluation and selection are very important, which relate to multiple factors, such as reliability;security, service time;the lifeeyele, etc. Results of application in torpedo design indicate that this method is feasible.