How can one construct a test that provides accurate measurements across the range of performance levels while providing adequate coverage of all of the critical areas of the domain, yet that is not unmanageably long? This paper discusses the approach taken in a linear test of academic English language, and how the transition to a computer-based test allowed for a design that better fit the demands of the test. It also describes the multi-stage adaptive approach that was devised. This approach allows for a test that covers a broad range of performance levels while including items that assess the language of the content areas as described in the English language development standards underpinning the test. The design also allows for a test that is closely tailored to the ability level of the English learner taking the test, and that therefore produces a more precise measure. The efficacy of the design in enhancing measurement of two versions of a high-stakes English language assessments is explored, and the implications of the results are discussed.