Ontologies are widely used in knowledge intensive applications such as health care, bioinformatics and the Semantic Web to provide shared languages for describing underlying domains. Ontologies are often modelled using description logics to provide succinct and expressive formalisms, well-defined semantics and efficient reasoning methods. The design and management of large complex ontologies in practice are often considered as life cycles, which raise the need for well-defined, efficient, and automated methods and tools to support ontology evolution over time. However, ontology evolution is difficult, as naive methods for ontology evolution may introduce incorrect semantics and even inconsistencies. Consequently, ontology evolution is poorly supported by the existing editing and reasoning tools. In this dissertation, we develop several new methods for ontology evolution, using description logics as underlying formalisms. We focus on two important ontology evolution operations: ontology reduction and ontology extension. These two operations correspond to the removal and addition of knowledge in ontologies, respectively, and both have important applications in ontology engineering. We define an ontology reduction operator, (model-based) forgetting, by adapting classical forgetting to description logics. Our forgetting operator is capable of eliminating terms from an ontology and reformulating the remaining axioms in a way that preserves the semantics of the remaining terms. In this sense, forgetting is more powerful than the existing module extraction approaches. We also introduce a parameterized definition of forgetting, called query-based forgetting, as a general framework that captures both forgetting and uniform interpolation in description logics.