Automatic system tests are frequently coupled to implementation details of the system under test. Such a tight coupling is problematic for a number of reasons: It prevents reuse of existing tests for multiple versions or variants of a system or for a number of systems in a system family and it makes the ability to construct executable tests for a system dependent on programming skills. In this paper we present an approach that decouples test specification and test execution by using system models and test models for the representation of systems under test and tests. We motivate the use of abstract test models and system models, introduce the relevant concepts of our approach and discuss the relationship to relevant fields.