Abstract: A study on the demographics of drunk drivers is conducive to improve the pertinence of traffic safety education.The data of 1 072 drunk drivers concerning four demographic variables, including age, gender, household registration, and educational level, is collected in a city within a one-year period.Afterwards, a log-linear model is developed to analyse the data.The results show that: ① Drunk drivers have significant differences on all the four variables;the male, the elder, the non-local, and the highly educated drivers are more likely to be involved in drunk driving.It finds that the non-local are more involved in drunk driving than the local, which is contrary to current studies and intuitive feelings.Therefore, the local have higher safety awareness than the non-local when it comes to drunk driving;② The interaction effects between two of the four variables are significant (p<0.01);the interaction effects between gender and age, gender and household registration are positive, while the other interaction effects are negative.Educational level adjusts the impacts of age, gender, and household registration.Age adjusts impacts of household registration on drink driving.The adjustment can explain the cause of drunk driving to a certain extent.Finally, based on this study, counter-measures about how to improve traffic safety education on drunk driving are discussed.