In software testing, GUI test cases are software functions that test the quality of the application under test (AUT) only passing through the graphical user interface (GUI). GUI test cases are well-known to be fragile, namely, they do not pass anymore as soon as an element in the GUI gets modified. The thesis, therefore, aims at investigating the sources of fragility in GUI-based test cases and how to decrease it. First, it is presented the overall background of software testing with particular attention to web applications. Then, a process of statistical collection and analysis on real projects aims at giving an objective overview of how testers face fragility and the forms under which fragility appears.