In situations such as multicriteria decision making or performance appraisal, linguistic values such as good, poor, or fair are usually used to represent results of evaluation. To grasp the vagueness of linguistic values, fuzzy numbers are introduced to represent uncertainties. Usually, input and output of the evaluation system for the fuzzy environment are fuzzy numbers. However, if the final results of the overall evaluation are represented by fuzzy numbers, it would be difficult to understand for ordinary people. We propose a linguistic computation model for performance evaluation, which takes linguistic values as input, computes on linguistic values and outputs results in terms of linguistic values.