The growing propensity to diagnose AD in individuals with very mild cognitive impairment increases the danger of false-positive diagnostic errors. Unfortunately, there is little systematically acquired information about the accuracy of the AD diagnosis in very mildly impaired patients. To determine the accuracy of the diagnosis of AD in very mildly impaired patients and to identify objective measures that effectively distinguish these patients from elderly normal controls (NC). Consecutive patients with Mini-Mental State Examination scores of > or = 24 who received a clinical diagnosis of AD were evaluated annually for at least 3 years. The initial diagnosis was verified or refuted by autopsy or by information obtained in subsequent evaluations. Initial neuropsychological test scores of verified AD patients were compared with those of NC subjects to identify effective diagnostic measures. The diagnosis of AD was confirmed in 98 of 110 (89%) very mildly impaired patients (33/36 by autopsy, 65/74 by disease progression). The diagnosis was inaccurate in 12 patients (11%): Seven were subsequently diagnosed with other neurologic disorders, and five were ultimately found to be normal. Neuropsychological measures of delayed recall, verbal fluency, and global cognitive status (i.e., Mattis Dementia Rating Scale) provided excellent sensitivity (> or = 96%) and specificity (> or = 93%) for differentiating between very mildly impaired AD patients and NC subjects. When comprehensive assessment procedures are employed, AD can be diagnosed with reasonably high accuracy in very mildly impaired individuals. However, the dementia evaluation should be repeated after approximately 1 year to ensure the accuracy of the initial diagnosis.