The role of computed tomography (CT) of the head in evaluating patients with dementing illnesses remains a controversial issue. Several prediction rules to guide the selective application of CT in the evaluation of dementia have recently been proposed in the medical literature. The present authors examine the value of four such rules through a validation study performed in an outpatient geriatric assessment unit. The rules were assessed in terms of their diagnostic sensitivities, specificities, misclassification rates, and information contents. Prediction rule sensitivities ranged from 12.5% to 87.5%, specificities from 37.2% to 77.9%, and misclassification rates from 23.5% to 60.8%. Of the four prediction rules examined, one emerged as significantly more sensitive than the others, and also served to reduce diagnostic uncertainty a full order of magnitude more than the others, as determined by an information content analysis. Disadvantages to this rule, however, were found in its more complex nature and the assessment of a very high rate of misclassification. Through a critique of existing strategies, this study purports to determine the potential for establishing a useful clinical prediction rule to guide selective CT scanning in the diagnostic evaluation of dementia.