Service-oriented systems (SOS) have recently emerged as context-independent component-based systems. In contrast to components, services can be created, invoked, composed and destroyed at run-time. Services are assumed to be platform independent and available for use within heterogeneous applications. One of the main assets in SOS is service composability. It allows the development of composite services with the main goal of reusable functionality provided by existing services in a low cost and rapid development process at run-time. However, in such distributed systems it becomes difficult to guarantee the quality of services (QoS), both in isolation, as well as of the newly created service compositions. Means of checking correctness of service composition can enable optimization w.r.t. the function and resource-usage of composed services, as well as provide a higher degree of QoS assurance of a service composition. To accomplish such goals, we employ model-checking technique for both single and composed services. The verification eventually provides necessaryinformation about QoS, already at early development stage.This thesis presents the research that we have been carrying out, on developing of methods and tools for specification, modeling, and formal analysis of services and service compositions in SOS. In this work, we first show how to formally check QoS in terms of performance and reliability for formallyspecified component-based systems (CBS). Next, we outline the commonalities and differences between SOS and CBS. Third, we develop constructs for the formal description of services using the resource-aware timed behavioral language called REMES, including development of language to support service compositions. At last, we show how to check service and service composition(functional, timing and resource-wise) correctness by employing the strongest post condition semantics. For less complex services and service compositions we choose to prove correctness using Hoare triples and the guarded command language. In case of complex services described as priced timed automata(PTA), we prove correctness via algorithmic computation of strongest post-condition of PTA. / Q-ImPreSS