The Confusion Assessment Method (CAM) is an easy, four-step algorithmic diagnostic test developed to detect delirium. To determine how sensitive and specific the CAM is in diagnosing delirium when compared with fully operationalized criteria of delirium according to the Diagnostic and Statistical Manual of Mental Disorders (DSM) editions III, III revised, and IV, and the International Classification of Diseases (ICD) 10th edition. A cross-sectional study with blinded assessments was performed on consecutive elderly patients (>70 years) (n=81) in two acute geriatric hospitals in Helsinki, Finland. The sensitivity, specificity, likelihood ratios, and positive and negative predictive values of CAM were assessed with the DSM-III, DSM-III-R, DSM-IV, and ICD-10 criteria of delirium used as reference standards. Sensitivity rates of the CAM were proved to be only moderate (0.81-0.86) against all formal criteria of delirium. The specificity rates were lower (0.63-0.84), and far less than reported in previous studies using global assessment of the reference standard. Instead of the DSM-III-R, from which it is derived, the CAM seems more concordant with the DSM-IV criteria of delirium. The likelihood ratio for a positive CAM test was 5.06 and for a negative test 0.23, when compared with the DSM-IV. The CAM seems to be an acceptable screening instrument for delirium, but the diagnosis should be ensured according to the formal criteria of delirium, preferably by the DSM-IV. Copyright 2002 John Wiley & Sons, Ltd.