Random-digit dialing procedures were used in association with computer-assisted telephone interviewing in Erie County (N.Y.) in 1992. The participants were categorized by their perceptions of the risk for drunk- driving arrest into a high probability group (at least 1-in- 10 chance for arrest) and a low probability group (a 1-in- 100 chance or less). Results revealed that men in the high probability group were significantly heavier drinkers than those in the low probability group. Women who drank heavily in a single sitting also tended to express a higher perceived risk of arrest. Age differences were also found. Findings suggested that drunk-driving prevention efforts must be directed at specific groups and that for some groups, such as heavier drinkers, efforts other than deterrence are needed. Tables, figure, note, and 18 references (Author abstract modified)