OBJECTIVE: To investigate the agreement of physician judgments of capacity to consent to treatment for normal and demented older adults. DESIGN: Subjects were individually administered a standardized consent capacity interview. Physicians viewed videotapes of these interviews and made judgments of capacity to consent to treatment. SETTING: University medical center. PARTICIPANTS: Subjects assessed for competency (N = 45) were 16 normal older controls and 29 patients with mild Alzheimer's disease (AD). Five medical center physicians with experience assessing the competency of dementia patients were recruited from the specialties of geriatric psychiatry, geriatric medicine, and neurology. MEASUREMENTS: Subjects were videotaped responding to a standardized consent capacity interview (SCCI) designed to evaluate capacity to consent to treatment. Study physicians blinded to subject diagnosis individually viewed each SCCI videotape and made a judgment of competent or incompetent to consent. Agreement of physician judgments was evaluated using percentage agreement, kappa, and logistic regression. RESULTS: Competency judgments of physicians showed high agreement for controls but low agreement for AD patients. Physicians as a group achieved 98% judgment agreement for the controls but only 56% judgment agreement for the mild AD patients. The physician group kappa for controls was 1.00 (P < .0001) and differed significantly (P < .0001) from the physician group kappa of .14 (P = .44) for AD patients, indicative of a real difference in the ability of the study physicians to judge consistently competency across the two groups. Similarly, logistic regression analysis showed significant variability in physician judgments for the AD group (X2 = 63.8, P < .0001) but not for the control group (X2 = 4.1, P = 1.00). Within the AD group, pairwise analyses revealed significant judgment disagreement (P < .01) for seven of the 10 physician pairs. CONCLUSIONS: Experienced physicians demonstrated significant agreement judging the competency of older controls, but significant disagreement judging the competency of mild AD patients. Physician judgment currently does not represent a gold standard for determining consent capacity of dementia patients. Standardized assessment instruments and clinical training may improve the agreement and accuracy of physician competency judgments.