Subjects were tested on general knowledge questions. They had to give their answer to each question and to state their level of confidence in its correctness. This was done under four conditions: by a paper and pencil test with and without feedback; and by computerized testing, again, with and without feedback. All in all, subjects demonstrated overconfidence in their knowledge under all conditions. However, feedback by a computerized system was effective in reducing the overconfidence level. The implications of this finding to the domain of computer-based educational systems is discussed.