A prospective, descriptive study was used to assess the diagnostic validity of the Confusion Assessment Method (CAM) administered at the bedside by nurses in daily practice. Two different scoring methods of the CAM (the specific (SPEC) and sensitive (SENS) methods) were compared with a criterion standard (CAM completed by trained research nurses). During a 5-month period, all patients consecutively admitted to an acute geriatric ward of the University Hospitals of Leuven (Belgium) were enrolled in the study. The 258 elderly inpatients who were included underwent 641 paired but independent ratings of delirium by bedside and trained research nurses. Delirium was identified in 36 of the 258 patients (14%) or in 42 of the 641 paired observations (6.5%). The SENS method of the CAM algorithm as administered by bedside nurses had the greatest diagnostic accuracy, with 66.7% sensitivity and 90.7% specificity; the SPEC method had 23.8% sensitivity and 97.7% specificity. Bedside nurses had difficulties recognizing the features of acute onset, fluctuation, and altered level of consciousness. For both scoring methods, bedside nurses had difficulties with the identification of elderly patients with delirium but succeeded in diagnosing correctly those patients without delirium in more than 90% of observations. Given these results, additional education about delirium with special attention to guided training of bedside nurses in the use of an assessment strategy such as the CAM for the recognition of delirium symptoms is warranted.