OBJECTIVES. Sources of measurement error in assessing smoking status are examined. METHODS. The Lung Health Study, a randomized trial in 10 clinical centers, includes 3923 participants in a smoking cessation program and 1964 usual care participants. Smoking at first annual follow-up was assessed by salivary cotinine, expired air carbon monoxide, and self-report. Each of these measures is known to contain some error. Sensitivity and specificity were calculated by comparing a biochemical measure with self-report to produce an undifferentiated estimate of error. Classification error rates due to imprecision of the biochemical measures and to the error in self-report were estimated separately. RESULTS. For cotinine compared with self-report, the sensitivity was 99.0% and the specificity 91.5%. For carbon monoxide compared with self-report, the sensitivity was 93.7% and the specificity 87.2%. The classification error attributed to self-report, estimated by comparing the results from intervention and control groups, was associated with the responses of 3% and 5% of participants, indicating a small but significant bias toward a socially desirable response. CONCLUSIONS. In absolute terms in these data, both types of error were small.