The present study compared the effects of an errorless stimulus shaping procedure to an errorful fluency based procedure for teaching difficult letter sound discriminations using a counterbalanced multielement experimental design. For 2 participants, letters fsteai were taught using the errorless procedure and letters bpdvou were taught using the errorful procedure. For the other 2 participants the conditions were reversed. All participants had considerably fewer errors and fewer trials to criterion with the errorless than with the errorful procedure. Tests of retention and generalization indicate that the errorful procedure generalized and was retained at a higher frequency than the errorless procedure. For 3 participants preference for the errorless procedure over the errorful procedure was demonstrated; whereas, the fourth participant demonstrated preference for the errorful procedure.