Large and distributed access control systems are increasingly common, for example in health care. In such settings, access control policies may become very complex, thus complicating correct and efficient adminstration of the access control system. Despite being one of the most widely used access control standards, RBAC does not include an administration model for distributed systems. In this paper we fill this gap. We present a model for the administration of RBAC in a distributed system and propose an administration procedure supporting the principle that different systems protect different sets of objects. We demonstrate that our procedure fulfills the formal requirements deriving from safety and availability, and we show how it can be translated to a practical implementation. Finally, we show how our model can be extended with multiple decentralized administrative systems.