This study aimed to determine whether the exclusive positive conditional relations established by the matching-to-sample (MTS) procedure are sufficient for equivalence class formation, or whether the negative conditional relations established with stimuli of alternative classes are necessary for it. In Experiment 1, two 3-choice MTS procedures were compared regarding equivalence class formation. The standard MTS procedure, where negative relations among stimuli of alternative classes are trained, was compared to an altered MTS procedure, where negative relations with stimuli that were not positive to any sample were trained. In Experiment 2, the positive and negative control patterns established by the standard and altered MTS procedures were assessed. Experiment 3 compared 2 further variations: (a) training only 1 negative relation with stimuli of alternative classes in each training trial type (semi-standard MTS procedure) or (b) varying the negative stimuli that did not belong to any class (varied-altered MTS procedure). The overall results indicate that for participants demonstrating high positive conditional baseline relations and high negative relations to stimuli from alternative classes, the probability of equivalence class formation was high, but when participants showed only high positive conditional baseline relations, the probability of equivalence class formation was very low. All main theories of equivalence class formation have difficulty accounting for these results, and an account based on a learning history of classifying behavior is offered. © 2015, Association for Behavior Analysis International.