Two tailored testing models, specifying procedures by which the correctness of examinees' responses to a fixed number of test items are predicted by presenting as few items as possible to the examinee, were compared for their efficiency. The models differ in that one requires reconsideration of each prediction whenever additional information is obtained, and the other does not. The results of computer simulations indicated that the testing procedures were more efficient when larger data bases were used. Losses incurred with small data bases were less than the losses that would be obtained if every item were presented to each examinee. (Author)