Police-reported crash data are rarely used to investigate safety belt use and its predictors, even though these data have a number of advantages over data collected in roadside surveys. It has been widely recognized that motorists tend to over-report their safety belt use to police when mandatory belt use becomes law. In this paper, we use a logistic regression model that allows for misclassification errors in outcome variable to examine predictors of safety belt use among crash-involved drivers and front seat passengers. Our analysis shows significant associations between occupant characteristics, driving circumstances, and safety belt use. Alcohol involvement has the strongest negative association with safety belt use, but this association would be considerably underestimated without adjusting for the over-reporting of safety belt use in police-reported crash data. The adjusted belt use rate among front seat occupants with at least nonincapacitating injuries is about 81%, compared to 90% in police-reported crash data.