Empirical evidence is presented related to the effects of using a stratified sampling of items in multiple matrix sampling on the accuracy of estimates of the population mean. Data were obtained from a sample of 600 high school students for a 36-item mathematics test and a 40-item vocabulary test, both subtests of the Iowa Tests of Educational Development. The results indicate that a stratified sampling of items, either by item difficulty level or by item discriminating ability, does not consistently yield more accurate estimates of the population mean than does simple random sampling. (Author/CTM)