Most of the time a large software system implies a complex architecture. However, at some point of the system’s execution, its components are not necessarily all running. Indeed, some components may not be concerned by a given use case, and therefore they do not consume/use or register the declared services. Thus, these architectural elements (components and their services) represent a “noise” in the architecture model of the system. Their elimination from the architecture model may greatly reduce its complexity, and consequently helps developers in their maintenance tasks. In our work, we argue that a large service-oriented system has, not only one, but several architectures, which are specific to its runtime use cases. Indeed, each architecture reflects the services, and thereby the components, which are really useful for a given use case. In this paper, we present an approach for recovering such use case specific architectures of service-oriented systems. Architectures are recovered both through a source code analysis and by querying the runtime environment and the service registry. The first built architecture (the core architecture) is composed of the components that are present in all the use cases. Then, depending on a particular use case, this core architecture will be enriched with only the needed components.