In this paper , a method of building an evaluation system with learning ability is presented. The evalua2 tion system is realized based on the decision2makers’knowledge and the fuzzy neural networks , and suitable for the problems in which most variables are linguistic variables. Samples used in the developed system are mined from de2 cision2makers’, and the corresponding membership functions are obtained by concerning with the preferences of the decision2makers’. The developed evaluation system is capable of using the previous decision experience and extract2 ing the knowledge from real decision examples. By training the fuzzy neural networks , the same evaluating results as the decision2makers’can be obtained.