We evaluated the accuracy of the field sobriety assessment of police officers following a motor vehicle crash. Using linked data from a statewide traffic crash database and the trauma center registry, the results of a field sobriety evaluation for each driver were compared with a blood alcohol concentration (BAC) drawn in the emergency department. BAC values served as the "gold standard" in which drivers with values over 0.1 mg/dl were classified as intoxicated. The subjects were drivers over age 15 years who were admitted or died at Level I trauma center as a results of motor vehicle crash injuries during 1986-1993. A total of 1336 subjects had both a recorded BAC and a police sobriety assessment. Seventy percent of subjects were male and 79% under the age of 35 years. Overall, 40% of all subjects were judged by police to have been impaired, based on the scene sobriety assessment. Among all drivers, the field assessment had a sensitivity of 91%, a specificity of 90% and a predictive value positive of 85%. After excluding the 419 drivers which had breath testing as part of their assessment, the field assessment had a sensitivity of 74% and a specificity of 97%. The sensitivity of the field assessment did not vary appreciably by gender but was lower among older drivers, and higher among severely injured drivers and those involved in weekend and nighttime crashes. Police officers in this sample appear to recognize drunk driving with a high degree of accuracy when investigating crashes in which the driver is transported to a trauma center.