Objectives: To determine the reliability, validity, and responsiveness to change of AUDIT (Alcohol Use Disorders Identification Test) questions 1 to 3 about alcohol consumption in a primary care setting. Patients: Randomly selected, male general medical patients (n= 441) from three VA Medical Centers, who had 5 or more drinks containing alcohol in the past year and were willing to be interviewed about their health habits. Measures: Three self-administered AUDIT consumption questions were compared with a telephone-administered version of the trilevel World Health Organization interview about alcohol consumption. Results: Of 393 eligible patients, 264 (67%) completed interviews. Test-retest reliability—Correlations between baseline and repeat measures 3 months later for four dimensions of consumption according to the AUDIT, ranged from 0.65 to 0.85, among patients who indicated they had not changed their drinking (Kendall's Tau-b). Criterion validity—Correlations between AUDIT and interview for four dimensions of alcohol consumption ranged from 0.47 to 0.66 (Kendall's Tau-b). Discriminative validity—The AUDIT questions were specific (90 to 93%), but only moderately sensitive (54 to 79%), for corresponding criteria for heavy drinking. Responsiveness to change—The AUDIT consumption questions had a Guyatt responsiveness statistic of 1.04 for detecting a change of 7 drinks/week, suggesting excellent responsiveness to change. Conclusions: AUDIT questions 1 to 3 demonstrate moderate to good validity, but excellent reliability and responsiveness to change. Although they often underestimate heavy alcohol consumption according to interview, they performed adequately to be used as a proxy measure of consumption in a clinical trial of heavy drinkers in this population.