Summary: In this thesis we analyze refinement operators that compute sets of specializations or generalizations of clauses. These operators are used to learn theories, which may be concept descriptions in the form of logic programs or sets of integrity constraints that are valid in a database. The research area of this work is Inductive Logic Programming (ILP), the intersection of inductive learning and logic programming. ILP inherits its main goal from inductive learning, namely to learn theories from examples. From logic programming, it inherits its representation formalism: (Horn-)clause logic.