The reversed association theory (Dunn & Kirsner, 1988) provides a powerful procedure for studying the link between cognitive processes and task performance. It helps find whether two behavioral tasks involve the same or different cognitive processes. However, this theory has not been fully utilized. Finding reversed association requires a large, single study consisting of at least six independent manipulations. Furthermore, the statistical procedure to verify reversed association has not been fully developed. This article presents practical solutions for these problems by investigating the recent controversy over categorization and similarity judgments. First, a contrast analysis is illustrated in a case study to statistically verify reversed association. Second, empirical experiments and computer simulations are presented to verify the reliability of the reversed association test. Combined together, this study reveals that there is a non-monotonic relationship (i.e., reversed association) in performance for categorization and similarity judgment tasks.