Investigators have reported poor recognition of dementia by primary physicians. For this reason, mental status examinations were performed on 72 demented and 144 nondemented medical inpatients to assess the sensitivity, specificity, and predictive value of components of this examination in the diagnosis of dementia. Sensitivity of individual level-of-orientation items was low (15.3 to 56.9%), though specificity was high (91.7 to 100%). Sensitivity of several nonorientation items was high (80.6 to 100%), though specificity was low. A multivariate discriminant equation using both orientation and nonorientation items achieved high sensitivity (89.6% test cases, 87.5% validation cases) and specificity (78.1% test cases, 87.5% validation cases). Adding the easily obtained patient characteristic of age to the equation further increased sensitivity (95.8%, 91.3%), while maintaining specificity (82.3%, 85.4%). From these results, it is concluded that the level-of-orientation screening examination used by clinicians to detect dementia is unacceptably insensitive. In contrast, a composite decision rule including nonorientation items achieves high sensitivity with relatively high specificity.