Model-based testing of state based systems is known to be able to spot non-conformance issues. However, up to half of these issues appear to be errors in the model rather than in the system under test. Errors in the specification at least hamper the prompt delivery of the software, so it is worthwhile to invest in the quality of the specification. Worse, errors in the specification that are also present in the system under test cannot be detected by model-based testing. In this paper we show how very desirable properties of specifications can be checked by systematic automated testing of the specifications themselves. We show how useful properties of specifications can be found by generalization of incorrect transitions encountered in simulation of the model.