Various models which are based on fuzzy systems have been successfully used in many application areas including but not limited to forecasting, optimization, clustering, and modeling. An important issue is to evaluate the performance of these models. In general, a performance measure is calculated based on the difference between the defuzzified output values and corresponding desired values. Although fuzzy inference is performed over fuzzy sets or numbers, performance measure is calculated over numerical values in the literature. Thus, the membership values are ignored. We suggest that a performance measure which takes membership values into account would be a better criterion. In this study, a novel membership value based performance measure which takes membership values into consideration is proposed to evaluate models based on fuzzy systems.