The choice of design parameters (e.g., number of subtests, number of items per subtest, and number of examinees per subtest) can be controlled by the test constructor in a multiple matrix sampling evaluation. The purpose of this study was to determine empirically which combination of the above parameters produces the smallest standard errors of the estimated mean and variance when certain assumptions are violated. The results indicate that multiple matrix sampling is a robust procedure even though proposed guidelines which should be followed to minimize the standard error terms are ignored. (Author)