This experiment explored how training modality influenced the formation of three-node, five-member equivalence classes during the simultaneous protocol by 23 college students. The baseline relations were established in one of three ways: concurrently (CONC; i.e., all together), serially (SER; i.e., one after another)—both on a trial and error basis—or serially and “errorlessly” by use of a constructed response matching to sample procedure (CRMTS). After training, class formation was assessed with test blocks that contained all baseline relations trials and probe trials for symmetry, transitivity, and equivalence. Error percentages during the acquisition of the baseline relations was highest during concurrent training, lower during serial training, and lowest during constructed response training. Yet, a similar percentage of participants formed classes in each training condition. Thus, the likelihood of equivalence class formation under these variants of the simultaneous protocol was not influenced by training modality or prevalence of errors during acquisition. In addition, only some of the transient nonclass-indicative stimulus control topographies that emerged during training resurged during testing. These results challenge predictions regarding class formation posited by stimulus control topography coherence theory.