When performing large scale evaluations (e.g., on a state-wide or national level) it may not be possible to administer all items in the item universe to all respondents in the subject population. One method which has been proposed to sample both items and respondents is multiple matrix sampling (MMS) in which a sample of the items is administered to a sample of subjects. In this study, the rules for constructing partially balanced incomplete block (PBIB) designs were followed in assigned items to subtests. Estimates of the population mean and variance were obtained and the results were compared with the estimates of the mean and variance obtained from using simple random sampling to construct subtests. In all comparisons between the two methods of assigning items to subtests, the number of subtests, the number of items per subtest and the total number of observations was held constant. The findings suggest that use of the PBIB designs may result in slightly less variable estimates of the mean, but simple random sampling will result in slightly less variable estimates of the variance. The primary conclusion is that the gain in precision which may occur when using PBIB designs is probably not worth the added effort and expense. (Author)