This paper investigated differences in the item selection performance of the following techniques: item-remainder correlation technique, principal component analysis technique, backward eliminating technique using item-remainder correlation, backward eliminating technique using principal component analysis, forward including technique using the ten Berge-Zegers μ2 coefficient as the internal consistency, and forward including technique using the Loevinger's H coefficient as the internal consistency. Using 3000 subject-item response matrices generated by artificial and real item response patterns of an achievement test, the simulation study showed that the forward including techniques were more appropriate to maximize the internal consistency of test than the other techniques, and this was true of several item pools. Practical attention to the forward including techniques and the other techniques were more furtherly discussed.