It has been suggested that the keystone-species concept should be dropped from ecology and conservation, primarily because the concept is poorly defined. This prompted Power et al. (1996) to refine the definition: keystone species have large effects on community structure or ecosystem function (i.e., high overall importance), and this effect should be large relative to abundance (i.e., high community importance). Using prairie dogs (Cynomys spp.) as an example, I review operational and conceptual difficulties encountered in applying this definition. As applied to prairie dogs, the implicit assumption that overall importance is a linear function of abundance is invalid. In addition, community importance is sensitive to abundance levels, the definition of community, and sampling scale. These problems arise largely from the equation for community importance, as used in conjunction with removal experiments at single abundance levels. I suggest that we shift from the current emphasis on the dualism between keystone and nonkeystone species and instead examine how overall and community importance vary (1) with abundance, (2) across spatial and temporal scales, and (3) under diverse ecological conditions. In addition, I propose that a third criterion be incorporated into the definition: keystone species perform roles not performed by other species or processes. Examination of how these factors vary among populations of keystone species should help identify the factors contributing to, or limiting, keystone-level functions, thereby increasing the usefulness of the keystone-species concept in ecology and conservation. Although the quantitative framework of Power et al. falls short of being fully operational, my conceptual guidelines may improve the usefulness of the keystone-species concept. Careful attention to the factors that limit keystone function will help avoid misplaced emphasis on keystone species at the expense of other species.