Many organisations in the safety-critical domain are service-oriented, fundamentally centred on critical services provided by systems and operators. Increasingly, these services rely on software-intensive systems, e.g. medical health informatics and air traffic control, for improving the different aspects of industrial practice, e.g. enhancing efficiency through automation and safety through smart alarm systems. However, many services are categorised as high risk and as such it is vital to analyse the ways in which the software-based systems can contribute to unintentional harm and potentially compromise safety. This thesis defines an approach to modelling and analysing Service-Oriented Architectures (SOAs) used in the safety-critical domain, with emphasis on identifying and classifying potential hazardous behaviour. The approach also provides a systematic and reusable basis for defining how the safety case for these SOAs can be developed in a modular manner. The approach is tool-supported and is evaluated through two case studies, from the healthcare and oil and gas domains, and industrial review.