Objectives: To assess validity and interrater reliability of an operationalized German version of the Confusion Assessment Method (CAM) in geriatric patients with comorbid dementia and high delirium risk. Design: Prospective cross-sectional cohort study with double CAM assessment by a medical and nonmedical rater. Setting: Random sample of frail, cognitively impaired elderly with acute disease requiring hospital care. Participants: A total of 39 frail elderly, mean age 83 ± 7 years, 72% (n = 28) female, with cognitive impairments, a high prevalence of dementia (86%, n = 33) and a significant risk of delirium. Of these, 13 revealed delirium, which was superimposed on dementia in 11. Measurements: A translated and operationalized version of the CAM was validated against a neuropsychiatric and geriatric consensus reference standard based on DSM-IV. Additional measures included the Short Portable Mental Status Questionnaire, the Mini-Mental State Examination and the Delirium Index for cognitive impairment severity, the Informant Questionnaire on Cognitive Decline for dementia diagnosis and the Barthel Index, illness severity (Cumulative Illness Rating Scale) and medication. Results: Delirium was correctly detected by CAM algorithm in 10 out of 13 delirious patients resulting in a high sensitivity of 0.77 and a specificity of 0.96–1.00 for both raters. Likelihood ratio revealed an almost 20-fold risk of delirium with positive CAM testing. Interrater reliability was excellent with a Cohen’s ĸ of 0.95 (CI 0.74–1.0) for the algorithm, single items’ ĸ values varied between 0.5 and 1. Conclusions: The German CAM is a reliable and valid measure of delirium, even in frail, acutely diseased elderly with concomitant dementia.