Conventional tests administered using paper-and-pencil to large numbers of examinees simultaneously have been a fixture of educational testing and measurement for many years. This testing strategy represents vastly reduced unit costs over tests administered individually, which were used during the early part of this century.However, interest in restoring some of the advantages of individualized testing has never completely disappeared. Turnbull suggested investigations in this direction in 1951 and coined the phrase tailored testing to describe this mode of test administration (Lord, 1980, p. 151). Possibilities for constructing individualized tests became likely with the advent of item response theory (IRT;Lord, 1952Lord, , 1980. In the 1960s, Lord (1970Lord ( , 1971a began to explore this application of IRT by investigating various item selection strategies borrowed from the bioassay field. Later work by Lord (1977Lord ( , 1980 and Weiss (1976, 1978) laid the foundation for the application of adaptive/tailored testing as an alternative to conventional testing.Adaptive tests are tests in which items are selected to be appropriate for the examinee-the test adapts to the examinee, usually by selecting items of appropriate difficulty. Computerized adaptive testing (CAT) has received increasing attention as a practical alternative to paper-and-pencil (Zara, 1990;Zara, Bosma, & Kaplan, 1987 (see Lord, 1970(see Lord, , 1971a(see Lord, , 1971b. Such investigations eventually led to IRT -based algorithms that were fast, efficient, and psychometrically sound. A review of the most frequently used algorithms is given in Wainer et al. (1990, chap. 5) and Lord (1980, chap. 9). The fundamental philosophy underlying these algorithms is as follows:1. An item is selected on some basis and administered to the examinee. (Ward, 1988) in which there are 10 to 15 item types. The same kind of control is used in the CAT-ASVAB (Segall, 1987). This type of content control has been called a constrained CAT (c-CAT) by Kingsbury & Zara (1989).A major disadvantage of this approach is that it assumes that the item features of interest partition the item pool into mutually exclusive subsets. Given the number of item features that may be of interest to test specialists, the number of mutually exclusive partitions can become very large and the number of items in each partition can become quite small. Moreover, incorporating considerations of overlap and item sets requires further partitioning by overlap group and by set, thereby further enlarging the number of mutually exclusive partitions. Wainer & Kiely (1987) hypothesized that the use of testlets could overcome these problems. They suggested that an adaptive test be constructed from testlets by using the testlet rather than an item as the branching point. They hypothesized that this would enable test specialists to enforce constraints on intrinsic item features, overlap, and item sets in the same manner as is currently done with conventional tests.Kingsbury & Zara (1991) compared ...