Services change over time, be it for internal improvements, be it for external requirements such as new legal regulations. For long running services, it may even be necessary to change a service while instances are actually running and interacting with other services. This problem is referred to as instance migration. We present a novel approach to the behavioral (service protocol) aspects of instance migration. We apply techniques for finitely characterizing the set of all correctly interacting partners to a given service. The approach assures that migration does not introduce behavioral problems with any running partner of the original service. Our technique scales up to services with thousands of states, including models of real WS-BPEL processes.