This paper presents an evaluation study that measures the effect of modifying feedback generality in an Intelligent Tutoring System (ITS) based on Student Models. A taxonomy of the tutor domain was used to group existing knowledge elements into plausible, more general, concepts. Existing student models were then used to measure the validity of these new concepts, demonstrating that at least some of these concepts appear to be more effective at capturing what the students learned than the original knowledge elements. We then trialled an experimental ITS that gave feedback at a higher level. The results suggest that it is feasible to use this approach to determine how feedback might be fine-tuned to better suit student learning, and hence that learning curves are a useful tool for mining student models.