Breath carbon monoxide (CO) is a convenient, widely used method for abstinence validation, with cutoffs of 8-10 ppm commonly employed. The goal of the present study was to determine an appropriate CO cutoff to differentiate nonsmokers and smokers within a large sample (N=374) of female prisoners incarcerated at a correctional facility in Virginia. Mean age of the population was 34.5 years, 49.2% were White, and 29% had less than a high school education. Smoking prevalence was 74.1% within the prison population. Examination of CO levels versus smoking self-report using a receiver operator characteristic (ROC) analysis revealed that a CO cutoff of 3 ppm resulted in the best sensitivity (98.1%) and specificity (95.8%). Overall ROC area under the curve was 99% (95% CI=98.2%-99.9%). This same cutoff was optimal for smoking subgroups including Black and light (<10 cigarettes/day) smokers. Results suggest that CO cutoffs higher than 3 ppm may misclassify some smokers as nonsmokers and underestimate the prevalence of smoking.