Linear discriminant analysis was used to construct a series of discriminant functions including subsets of demographic variables and Mini-Mental State Examination item responses for a case series and a population sample. A 9-item discriminant function including the variables for time orientation, recall, calculation, copying a figure, age, writing, 3-step command, naming, and race distinguished demented subjects from community controls with 91% sensitivity and 88% specificity. The same discriminant function classified Alzheimer's disease patients and controls with 96% sensitivity and 98% specificity. This discriminant function has been cross-validated as a potential screening instrument for Alzheimer's disease in a community-based sample.