Nowadays, using Service-Oriented Architectures (SOA) is spreading as a flexible architecture for developing dynamic enterprise systems. Due to the increasing need of high quality services in SOA, it is desirable to consider different Quality of Service (QoS) aspects in this architecture as security, availability, reliability, fault tolerance, etc. In this paper we investigate fault tolerance mechanisms for modeling services in service-oriented architecture. We propose a metamodel (formalized by a type graph) and some graph rules for monitoring services and their communications to detect faults. By defining additional graph rules as reconfiguration mechanisms, service requesters can be dynamically switched to a new service (with similar descriptions). To validate our proposal, we use our previous approach to model checking graph transformation using the Bogor model checker.