This paper is concerned with model-based testing of hybrid systems. The first result is an algorithm for test generation which enhances the coverage of critical trajectories by using a random walk. The second result is a framework for practical testing that includes a state estimator. When the state of a system under test cannot be directly observed, it is necessary to reconstruct the trajectory of the real system in order to produce a verdict whether the system violates a property. To do so, we integrate in our tester a hybrid observer, the goal of which is to provide an estimate for the current location and the continuous state of the system under test based on the information on the input and the output of the system.