Two experiments were conducted to examine the emergence of third-order equivalence classes. In Experiment 1, college students were trained in three successive phases to perform first-, second-, and third-order baseline conditional discriminations. In each phase, after attaining the baseline criterion, equivalence tests were conducted. Passing a test in a given phase was required for advancing to the next phase. Five of ten subjects showed performances consistent with third-order equivalence classes. To understand why higher-order relations sometimes did not emerge, the study was replicated. In Experiment 2, four subjects were instructed to talk aloud during all training and test phases. Two subjects showed third-order equivalence; the other two failed the second-order symmetry test and were unable to advance beyond Phase 2. A protocol analysis (Ericsson & Simon, 1980; Wulfert, Dougher, & Greenway, 1991) of their verbal responses revealed that their behavior was not controlled by the intended second-order conditional stimuli. When control by these stimuli was established through instructions, second-order, and in Phase 3 third-order, equivalence relations emerged. This research replicates previous studies on second-order equivalence and extends contextual control over emergent relations to the third order. The demonstration of complex emergent relations may have implications for a behavioral analysis of cognitive-verbal phenomena such as concept formation and classification.