Previous studies have investigated the viability of a go/no-go discrimination procedure (pREP) for generating stimulus equivalence. During pREP training, participants receive positive feedback for pressing a bar after the successive presentation of two same-class stimuli and for not pressing after the presentation of two different-class stimuli. During pREP testing, the procedures are the same but without feedback. The results of that research consistently indicated that pREP training (Ai[right arrow]Bi[right arrow]press, Ai[right arrow]Bj[right arrow]no press, Bi[right arrow]Ci[right arrow]press, Bi[right arrow]Cj[right arrow]no press) produces pREP symmetry (e.g., Bi[right arrow]Ai[right arrow]press, Bi[right arrow]Aj[right arrow]no press) but not pREP equivalence (Ci[right arrow]Ai[right arrow]press, C [right arrow]Aj[right arrow]no press) unless equivalence had already been demonstrated with match-to-sample (MTS), or the pREP was converted into a MTS task with the stimulus pairs as samples and the words SAME and DIFFERENT as comparisons. The present study examined if the efficacy of the pREP could be improved while leaving the go/no-go structure intact. The study consists of six experiments. The results indicated that the pREP had one, possibly two weaknesses. First, during testing, it permitted nondifferential responding. Second, during training, the contingencies permitted class ambiguity, that is, class formation based on pressing (Ai-Bi-Ci) or on not pressing (Ai-Bj-Ci). When measures were taken to discourage nondifferential responding (instruction to press on 50% of the trials, using a simple-to-complex test protocol) and to reduce ambiguity (feedback for pressing only), the pREP not only produced equivalence but also equivalence reversal in (almost) all participants. (Contains 6 tables.)