We report the results of mining student learning data from SlideTutor - a cognitive tutor in a medical diagnostic domain. The analysis was aimed at finding both individual learning patterns as well as common misconceptions that students possessed. We have discovered that indeed there are distinct learner stereotypes: hint-driven learners, failure-driven learners, and a mixed group of learners that cannot be attributed to either one of the above two types. We have also found that students often make similar mistakes confusing certain visual features and diagnostic hypotheses. Our goal is to reuse the discovered patterns to engineer cross-case pedagogic interventions, enhancing our current immediate feedback methods with higher-level pedagogic reasoning. This paper describes the data-mining activities and potential implications of the data for pedagogic design.