The cognitive assessment protocol of the Alzheimer's Disease Cooperative Study (ADCS) was designed to evaluate the reliability and validity of cognitive assessment measures that might be valuable additions to the Alzheimer's Disease Assessment Scale (ADAS) or other concise batteries used in antidementia drug trials. As part of an overall ADCS protocol to develop new instruments to be used in trials of treatments for Alzheimer's disease (AD), patients with mild to moderate AD and cognitively normal elderly were administered a battery of five tests at least three times over 1 year. The tests included word list learning with delayed free recall, a recognition memory test for faces, a series of letter and digit cancellation tests to measure concentration, tests of praxis, and a series of maze completion tasks designed to assess planning and executive function. A version of the digit cancellation task was reliable and sensitive to a broad range of dementia severity so that it could provide a useful addition to the present version of the ADAS. Performance on the word learning task with delayed recall and a subset of the mazes task were impaired even in mild AD, so these tasks may be useful in trials involving mild or at-risk subjects. Performances on the facial recognition task and on the praxis tasks were not related to dementia severity, so these tasks would not be useful to evaluate treatments. Therefore, the major outcome of this investigation was the identification of some potential additions to the present ADAS that extend both the cognitive domains and the range of symptom severity covered.