This study attempts to reconcile viewpoints on the role of errors in the design of computer-based training (CBT) systems. From one perspective, errors are detrimental to learning; from another, they can be beneficial. The results of an experiment are presented and discussed. When a CBT system is designed to allow learners to correct their own errors, they learn significantly more than when errors are prevented. When learners are allowed only to correct errors that others have made, they do not learn significantly more than when errors are prevented. Implications for the design of CBT systems and possibilities for future research are considered. (Contains 5 tables and 6 figures.)