Traditional evaluation methods do not actualize attribute reduction before evaluation, and evaluation process is too complex to new samples. This paper has studied attribute reduction and rule generation in comprehensive evaluation making use of the (S, T)fuzzy rough set. Attribute reduction is to optimize evaluating index and rule generation is to simplify calculation process, so a new method is provided to evaluation of new samples. At last, the paper gives an example of comprehensive evaluation to demonstrate the process of attribute reduction and rule generation.