We conducted a parametric analysis of treatment integrity errors during discrete-trial training and investigated the effects of three integrity conditions (0, 50, or 100 % errors of commission) on performance in the presence and absence of programmed errors. The presence of commission errors impaired acquisition for three of four participants. Following removal of commission errors from instruction, we observed a carry-over effect for only two participants suggesting that adverse effects are reversible with relatively small amounts of high-quality instruction.