Subjects were 18 police oficers employed by a midwestern university. The officers ranged in age from 22 to 50, with 6 months to 25 years police experience. Thirty-six students at the same university were randomly assigned a high or low dose of alcohol. Each police officer made three consecutive assessments of the intoxication level of one high dose student and one low student. Officers predicted the student's blood alcohol concentrations (BAC) while questioning the student about a hypothetical domestic dispute. For each student, the officer made a BAC judgement 30 seconds and 5 minutes after the interview began, and once again after it ended. Subjects were limited to 20 minutes with each student, each of whom was breath-tested before and after the interview. Subjects predicted sustantially the same BAC of 0.05 percent to 0.06 percent for students on both high and low alcohol doses. Actual BAC's for the high and low doses were about .15 percent and .075 percent respectively. Furthermore, predictions per subject did not change over the three assessments per student, although police confidence in their assessments increased. Police were not more confident in assessing the high dose students, however. Findings suggest that the failure of Bard and Zacker to find a strong relationship between alcohol use and assault resulted from police officers' inability to estimate alcohol intoxication reliably. Eleven references are included.