Many information systems developers face high cost in adopting service oriented architectures because of the high cost of locating appropriate services to customise and integrate into their system. This paper aims at reducing this cost by automating much of the composition and service selection effort. It illustrates the use of a Peer-Peer multi agent system (MAS) to facilitate service selection with multiple Quality of Service properties. The system will use semantic enrichment of services in order to facilitate their identification and composition. With semantic driven composition, services can be shared between teams of developers and across multiple organisations connected via the Internet. In this paper, we focus on a conceptual framework for peer selection with a preliminary mathematical model and a selection process, so as to enhance the P2P-based service coordination system and its components.