OBJECTIVE Jurisdictions with per se breath alcohol legislation rely heavily on breath test evidence in prosecuting drunk driving cases. Depending on other legal considerations, where subjects refuse the breath test, prosecution may be more difficult. The objective was to identify factors significantly associated with the risk of test refusal. This knowledge would be relevant for improving the compliance rate. METHODS A retrospective observational study evaluated drunken driving arrest records (n = 38,687) within Washington State during 2003 where breath tests were requested under implied consent legislation. The association was determined between the risk of test refusal and several categorical variables including: arresting agency, gender, race, age, accident involvement, repeat offense, pre-arrest breath test, driver's license and all possible two-way interactions. The association was quantified by odds ratios (OR) and 95% confidence intervals (CI) using a logistic regression model. RESULTS The overall refusal rate was 19.7%. The single main effect variable associated with the greatest reduction in risk of refusal was the performance of a pre-arrest breath test (OR = 0.31, 95% CI 0.30-0.33). No single main effect variables showed significant association with an increased risk of refusal. Several two-way interactions, however, were significant. Other possible predictor variables, not included in the analysis, are also discussed. CONCLUSIONS An arrested subject's decision to refuse the breath test is clearly a multivariate issue. Knowledge of these results should help in developing a jurisdictionally specific breath test compliance model by identifying the legal, demographic and procedural factors contributing most significantly to breath test refusal rates.