Association rules, a class of important regularities in databases, have proven very useful in practical applications, but association-rule-mining algorithms tend to produce huge numbers of rules, most of which are of no interest. Users have considerable difficulty manually analyzing so many rules to identify the truly interesting ones. To solve that problem, we have developed a new approach to help them find interesting rules (in particular, unexpected rules) from a set of discovered association rules. This interestingness analysis system (IAS) leverages the user's existing domain knowledge to analyze discovered associations and then rank discovered rules according to various interestingness criteria, such as conformity and various types of unexpectedness. This article describes how we have implemented this technique and used it successfully in a number of applications.