This paper presents a single project experiment on the fault revealing capabilities of model-based test sets. The tests are generated from UML statecharts and UML sequence diagrams. This experiment found that the statechart test sets did better at revealing unit level faults than the sequence diagram test sets, and the sequence diagram test sets did better at revealing integration level faults than the statechart test sets. The statecharts also resulted in more test cases than the sequence diagrams. The results show that model-based testing can be used to systematically generate test data and indicates that different UML models can play different roles in testing.