Most large-scale enterprise applications are currently built using component-based middleware platforms such as J2EE or .NET. Developers leverage enterprise services provided by such platforms to speed up development and increase the robustness of their applications. In addition, using a component-oriented development model brings benefits such as increased reusability and flexibility in integrating with third-party systems. In order to provide the required services, the application servers implementing the corresponding middleware specifications employ a complex run-time infrastructure that integrates with developer-written business logic. The resulting complexity of the execution environment in such systems makes it difficult for architects and developers to understand completely the implications of alternative design options over the resulting performance of the running system. They often make incorrect assumptions about the behaviour of the middleware, which may lead to design decisions that cause severe performance problems after the system has been deployed. This situation is aggravated by the fact that although application servers vary greatly in performance and capabilities, many advertise a similar set of features, making it difficult to choose the one that is the most appropriate for their task. The thesis presents a methodology and tool for approaching performance management in enterprise component-based systems. By leveraging the component platform infrastructure, the described solution can nonintrusively instrument running applications and extract performance statistics. The use of component meta-data for target analysis, together with standards-based implementation strategies, ensures the complete portability of the instrumentation solution across different application servers. Based on this instrumentation infrastructure, a complete performance management framework including modelling and performance prediction is proposed. Most instrumentation solutions exhibit static behaviour by targeting a specified set of components. For long running applications, a constant overhead profile is undesirable and typically, such a solution would only be used for the duration of a performance audit, sacrificing the benefits of constantly observing a production system in favour of a reduced performance impact. This is addressed in this thesis by proposing an adaptive approach to monitoring which uses execution models to target profiling operations dynamically on components that exhibit performance degradation; this ensures a negligible overhead when the target application performs as expected and a minimum impact when certain components under-perform. Experimental results obtained with the prototype tool demonstrate the feasibility of the approach in terms of induced overhead. The portable and extensible architecture yields a versatile and adaptive basic instrumentation facility for a variety of potential applications that need a flexible solution for monitoring long running enterprise applications.