This simulation study involved the design of two two-stage tests in which the routing tests and the second-stage measurement testlets took the form of a multidimensional knapsack problem with prespecified target informations and constraints to be enumerated using the algorithm of E. Balas. Two conventional tests of similar length to the two-stage tests were developed by random selection of items from the same item pool. Item responses of 1,000 simulees on 240 items were generated using the computer program DATAGEN based on the three-parameter item response model. Analysis of the degree in which the true abilities were recovered showed that the inaccuracies and root square mean errors for the two-stage designs were lower than those for their conventional counterparts. Fidelity coefficients were also higher and compared well with the results of previous adaptive test procedures. Three tables and three graphs accompany the discussion. (Author/SLD)