Flexible manufacturing systems often are organized into a cellular architecture for ease of operation. The formation of these cells sometimes has been treated as an extension of the conventional cell-formation problem. This paper argues that, owing to the existence of flexible routing and transfer capabilities, the cell-formation problem in FMSs should be treated as quite distinct from that in conventional manufacturing systems and shows that a flexibility-based procedure is apt for overcoming the deficiencies of earlier forays into this area. Manufacturing cell flexibility is defined as a composite of three flexibility measures: producibility, processivity, and transferability. The problem of cell formation is modeled as flexibility maximization, and a procedure is developed for the simultaneous formation of machine cells and part families, while heuristically maximizing within-cell flexibility.