Behavioral interventions that provide incentives contingent upon abstinence are effective addiction treatments. Nevertheless, these treatments often fail for individuals whose recent behaviors are very different from those reinforced. These hard-to-treat individuals may require shaping to achieve abstinence. We used percentile schedules to shape smokers' delivery of breath samples indicative of recent smoking abstinence (breath carbon monoxide (BCO) <4 ppm). Percentile schedules deliver incentives to current behaviors proximal to the target. Participants (N = 102) were assigned to treatments delivering incentives for breath COs at or below the 10th, 30th, 50th, or 70th percentile of recent breath COs. Each condition effectively ensured contact with available contingencies, and resulted in BCO <4 ppm in >90% of the 30th, 50th and 70th percentile groups versus 63% in the 10th percentile. The 30th, 50th and 70th percentiles were especially effective in a sub-sample of hard-to-treat participants who did not deliver a breath CO <4 ppm during an initial abstinence test or during a nine-visit baseline period, suggesting the value of shaping for this important sub-sample.