Localizing faults in programs is considered a demanding task. A lot of effort is usually spent in finding the root cause of a misbehavior and correcting the program such that it fulfills its intended behavior. The situation is even worse in case of end user programming like spreadsheet development where more or less complex spreadsheets are developed only with little knowledge in programming and also testing. In order to increase quality of spreadsheets and also efficiency of spreadsheet development, tools for testing and debugging support are highly required. In this paper, we focus on the latter and show that approaches originating from Artificial Intelligence can be adapted for (semi-) automated fault localization in spreadsheets in an interactive manner. In particular, we introduce abstract models that can be automatically obtained from spreadsheets enabling the computation of diagnoses within a fraction of a second. Besides the basic foundations, we discuss empirical results using artificial and real-world spreadsheet examples. Furthermore, we show that the abstract models have a similar accuracy to models of spreadsheets capturing their semantics.