Component technologies, such as J2EE and .NET have been extensively adopted for building complex enterprise applications. These technologies help address complex functionality and flexibility problems and reduce development and maintenance costs. Nonetheless, current component technologies provide little support for predicting and controlling the emerging performance of software systems that are assembled from distinct components. Static component testing and tuning procedures provide insufficient performance guarantees for components deployed and run in diverse assemblies, under unpredictable workloads and on different platforms. Often, there is no single component implementation or deployment configuration that can yield optimal performance in all possible conditions under which a component may run. Manually optimising and adapting complex applications to changes in their running environment is a costly and error-prone management task. The thesis presents a solution for automatically optimising the performance of component-based enterprise systems. The proposed approach is based on the alternate usage of multiple component variants with equivalent functional characteristics, each one optimized for a different execution environment. A management framework automatically administers the available redundant variants and adapts the system to external changes. The framework uses runtime monitoring data to detect performance anomalies and significant variations in the application's execution environment. It automatically adapts the application so as to use the optimal component configuration under the current running conditions. An automatic clustering mechanism analyses monitoring data and infers information on the components' performance characteristics. System administrators use decision policies to state high-level performance goals and configure system management processes. A framework prototype has been implemented and tested for automatically managing a J2EE application. Obtained results prove the framework's capability to successfully manage a software system without human intervention. The management overhead induced during normal system execution and through management operations indicate the framework's feasibility.