To identify the most effective method of screening for covert alcoholism Ewing's CAGE questionnaire was compared with several computer-assisted laboratory data profiles in a prospectively gathered, random sample of 915 adults admitted to a general hospital. Whether a subject was alcohol dependent (n = 244) or not (n = 671), as defined by DSM-III-R, was determined on the basis of a structured interview. The CAGE questionnaire was highly sensitive (76%) and specific (94%) for recognition of alcohol dependence (positive predictive power 87%). None of the discriminant laboratory functions gave recognition rates greater than chance alone. Until the sensitivities, specificities, and positive predictive powers of computer-assisted methods improve, brief interview alone remains the best screening method for general hospital populations.