Abstract. Web services are becoming progressively popular in the building of both inter-and intra-enterprise business processes. These processes are composed from existing Web services based on defined requirements. In collecting together the services for such a composition, developers can employ languages and standards for the Web that facilitate the automation of Web service discovery, execution, composition and interoperation. However, there is no guarantee that a composition of even very good services will always work. Mechanisms are being developed to monitor a composition and to detect and recover from faults automatically. A key factor in such self-healing is to know what faults to look for. If the nature of a fault is known, the system can suggest a suitable recovery mechanism sooner. This paper proposes a novel taxonomy that captures the possible failures that can arise in Web service composition, and classifies the faults that might cause them. The taxonomy covers physical, development and interaction faults that can cause a variety of observable failures in a system's normal operation. An important use of the taxonomy is identifying the faults that can be excluded when a failure occurs. Examples of using the taxonomy are presented.