Most of today's software users interact with the software through a graphical user interface (GUI). While GUIs have become ubiquitous, testing of GUIs has remained until recently, a neglected research area. Existing GUI testing techniques are extremely resource intensive primarily because GUIs have very large input spaces. This research proposes to advance the state-of-the-art in GUI testing by empirically studying GUI faults, interactions between GUI events, why certain event interactions lead to faults, and use the results of these studies to develop cost-effective model-based GUI testing techniques. The novel feature of this research will be a reduced model of the GUI's event-interaction space. The model will be derived automatically from the GUI; it will be used to automatically generate specialized GUI test cases that are effective at detecting GUI faults. The model will be extended to develop new test oracles, new coverage criteria for GUIs, and new regression testing techniques. Moreover, this research will empirically evaluate the developed techniques.