Mutation analysis was introduced in the seventies of the last century to assess the efficiency of a given test set. This approach makes use of specific mutation operators to systematically modify the source code of the system under test (SUT) to simulate typical programming errors. For each of the generated mutants, it checks whether or not the given test set detects the injected fault(s). The ratio of detected to undetected faults is finally used to assess the efficiency of the given test set. This thesis introduces a model-based mutation analysis. Instead of the source code of the SUT, its given model, which is assumed to be correct, is systematically modified. A test set is generated for each mutant using a model-based test generation algorithm, which is then applied to the SUT. In contrast to code-based mutation analysis the present model-based approach also enables the detection of "real," non-injected faults in the SUT. Another advantage is that the source code of the SUT is not needed, as the mutation analysis is performed on the model level. This eases the use in practice since the source code of the SUT is not always available. The beginning of the thesis explains the approach and syntactically introduces the basic mutation operators by means of directed graphs. To take the practical issues into account, well-known models are considered: event sequence graphs, finite-state machines, and statecharts. Using these models, several case studies identify and analyze the characteristics and advantages of the approach.