In this paper we propose a new approach far part families considering multiple attributes of parts in the design and operating stage of flexible manufacturing systems. We first represent the relationship of parts and the relative attributes using fuzzy membership function, AHP method and normalization. As a result, more realistic nonbinary data of the relationship is obtained. Then we group parts into part families based on the nonbinary data using fuzzy $alpha$-cut and new similarity coefficient method. The performance of our method is compared numerically with others.