Service-oriented architecture (SOA) is a popular industry standard that provides an elegant model to create and compose interoperable applications. In the case of sensor networks, SOA is considered to be a very attractive enabler for sensor programmability and self-integration. However, SOA cannot be used without additional mechanisms to enhance the reliability and availability of the sensors behind the services. In the absence of such mechanisms, sensor-based SOA could provide a false promise of service whose, quality and availability are highly questionable and cannot be guaranteed. In this paper, we propose a Virtual Sensor framework for service-oriented sensor networks with the aim of enhancing reliability, prolonging availability and enabling automatic service composability. We classify and model three useful virtual sensor concepts and present compensation algorithms for coping with sensor failure. To quantify the merit of our approach, we developed a Data Quality Indicator measure for expressing the perceived quality of virtual sensor readings.