The clinical diagnosis of delirium has traditionally been based on an assessment by one or more physicians. Because of the transient, ubiquitous, and fluctuating nature of the symptoms of delirium, however, this approach may be flawed. Therefore, we decided to compare diagnosis based on one assessment by a psychiatrist, diagnosis by a nurse clinician (using the Confusion Assessment Method [CAM] and multiple observation points), and diagnosis by consensus. The study subjects were 87 patients aged 65 and over who were admitted consecutively from the emergency department to the medical wards, and who scored 3 or more on the Short Portable Mental Status Questionnaire. All subjects were assessed independently by one of three psychiatrists (a chart review and clinical examination) and a nurse clinician (using the CAM and multiple observation points). A consensus conference, attended by the three psychiatrists and the nurse clinician, used all available information to reach a consensus diagnosis. Compared to the consensus diagnosis, the clinical diagnosis by a psychiatrist had a sensitivity of .73 (95% confidence interval [CI]: .61-.85), a specificity of .93 (95% CI: .79-1.0), and an agreement kappa coefficient of .58 (95% CI: .41-.74). The nurse clinician diagnosis had a sensitivity of .89 (95% CI: .81-.97), a specificity of 1.00, and an agreement kappa coefficient of .86 (95% CI: .75-.97). These results suggest that one clinical assessment by a psychiatrist may not be the best method for detecting and diagnosing delirium in the elderly. A consensus diagnosis or diagnosis by a trained rater (using the CAM and multiple observation points) may be more sensitive approaches.