Tobacco smoking is a well-documented risk factor for multiple illnesses including coronary heart disease, chronic obstructive pulmonary disease (COPD), stroke, and cancer [1] and is the leading cause of preventable deaths in the United States (U.S.) [2]. Despite these risks, the prevalence of tobacco use remains relatively high, making smoking an important potential confounder in epidemiologic studies. Ideally, detailed smoking history data are available to address confounding. However, even basic measures, namely ever/never or current smoking, cannot be directly assessed accurately in Medicare data. Claims contain diagnosis and procedure codes specific to smoking but only identify ~7% of past and current smokers [3], [4], [5]. Accordingly, although claims-based algorithms for smoking status have high specificity, sensitivity is limited [3], [4], [5], [6], [7], [8]. Claims-based data are detailed in many other respects, making them an important resource for researchers if this limitation can be overcome. We sought to develop and validate a claims-based algorithm to estimate the probability that a beneficiary had ever smoked. We hypothesized this algorithm would be sufficiently accurate for claims-based epidemiologic studies.