Using a computer-based Monte Carlo approach to generate item responses, the results of this study indicate that, when item discrimination indices are considered, item-examinee sampling procedures having the same number of observations have different standard errors in estimating both test mean and test variance. With certain types of tests, a single item-examinee sampling plan would not yield optimal, i.e., smallest standard error, estimates of both mu and sigma squared. That is, one sampling plan would be needed to optimally estimate mu and another to optimally estimate sigma squared. In addition, it was found that single exhaustion of the item set was sufficient for estimating both mu and sigma squared. (Author)