Adjustment of a cognitive test for an expected level of performance improves the discrimination between brain-diseased and healthy subjects. However, this improvement is subject to severe limitations and may be worthwhile only in clinical settings, where test results tend to be low regardless of disease status. The objective of this study was to provide an empirical demonstration of these principles, applied to the detection of dementia with the Mini-Mental State Examination (MMSE). The subjects, derived from a population-based sample, consisted of 36 cases of dementia (23 diagnosed shortly after testing and 13 at follow-up 1 year later) and 301 nondemented subjects defined by a negative follow-up diagnosis. A simulated group of 179 clinically suspect normals was obtained by selecting all cases with an MMSE score below 27. Adjustment was based on the Dutch version (DART) of the National Adult Reading Test (NART), which was highly correlated (0.53) with the MMSE score of nondemented subjects. The results were in accordance with the predictions. We conclude that adjustment is unlikely to improve case finding in representative samples, but can be profitable in clinical practice, where it will be especially helpful in ruling out cerebral disease.