Recent experience with the Computerized Adaptive Test (CAT) has raised a number of concerns about its practical applications. The concerns are principally involved with the concept of having the computer construct the test from a precalibrated item pool, and substituting statistical characteristics for the test developer's skills. Problems with CAT that can arise in practice are related to context effects (item location, cross-information, unbalanced context), lack of robustness, item difficutly ordering, and high technology factors. A variety of traditional options to circumvent these problems are explored, and a new methodology for solving them is offered. In order to use the testing efficiencies of CAT and the experience of test developers, this study proposes the bundling of items into "testlets." Items in testlets can be arranged hierarchically or linearly to yield increased efficiency, but the paths an examinee can follow are strictly determined. One potential drawback of the testlet approach is that the aggregate of items is treated as a unit, and it may be difficult to replace pieces of it independently of the others. (JAZ)