The researchers evaluated the Alcohol Use Disorders Identification Test (AUDIT), the first 3 questions of the AUDIT (AUDIT-C), the third AUDIT question (AUDIT-3), and quantity-frequency questions for identifying hazardous drinkers in a large primary care sample. Cross-sectional survey. Patients waiting for care at 12 primary care sites in western Pennsylvania from October 1995 to December 1997. Sensitivity, specificity, likelihood ratios, and predictive values for the AUDIT, AUDIT-C, and AUDIT-3. A total of 13,438 patients were surveyed. Compared with a quantity-frequency definition of hazardous drinking (> or =16 drinks/week for men and > or =12 drinks/week for women), the AUDIT, AUDIT-C, and AUDIT-3 had areas under the receiver-operating characteristic curves (AUROC) of 0.940, 0.949, and 0.871, respectively. The AUROCs of the AUDIT and AUDIT-C were significantly different (P=.004). The AUROCs of the AUDIT-C (P<.001) and AUDIT (P <.001) were significantly larger than the AUDIT-3. When compared with a positive AUDIT score of 8 or higher, the AUDIT-C (score > or =3) and the AUDIT-3 (score > or =1) were 94.9% and 99.6% sensitive and 68.8% and 51.1% specific in detecting individuals as hazardous drinkers. In a large primary care sample, a 3-question version of the AUDIT identified hazardous drinkers as well as the full AUDIT when such drinkers were defined by quantity-frequency criterion. This version of the AUDIT may be useful as an initial screen for assessing hazardous drinking behavior.