The study sought to determine possible sentencing thresholds and the optimum sentence length that produced the maximum deterrent effects. The participants were male drunk drivers in custody between October 1989 and June 1991. The sentences were grouped into 4 categories: (1) 90-120 days, (2) 121-186 days, (3) 187-366 days, and (4) 367 days and over. The analysis used multivariate statistical techniques to control for possible confounding background variables. Results revealed that sentence length exerted consistent deterrent effects on repeat drunk driving, even for chronic offenders. Shorter sentences were less effective in discouraging drunk driving recidivism, while sentences longer than 6 months did not produce additional benefits. The analysis concluded that specific deterrence and the principle of severity should receive greater attention and that policymakers should look to a more rational, calculated use of incarceration to reduce recidivism. Future research should use prospective designs and better specify the conditions and situations whereby specific deterrence can obtain optimal results. Notes and 30 references (Author abstract modified)