This paper presents the APRIORI-C algorithm, modifying the association rule learner APRIORI to learn classification rules. The algorithm achieves decreased time and space complexity, while still performing exhaustive search of the rule space. Other APRIORI-C improvements include feature subset selection and rule post-processing, leading to increased understandability of rules and increased accuracy in domains with unbalanced class distributions. In comparison with learners which use the covering approach, APRIORI-C is better suited for knowledge discovery since each APRIORI-C rule has high support and confidence.