We describe the persistence of discrepancies between biochemical measures of smoking and self-reported smoking status in a cohort of clinical trial participants across 5 years. The Lung Health Study, a randomized trial in 10 clinical centers in North America, enrolled 3923 participants in smoking intervention and 1964 in usual care in 1987 and 1988. Smoking status was assessed at baseline and at five annual follow-up visits by self-report, salivary cotinine and expired-air carbon monoxide. Compared to self-report, sensitivity and specificity of cotinine and carbon monoxide were similar across 5 years. Evidence of error in self-reports of quitting smoking persisted across 5 years, although it declined over time. Multivariate models confirmed that self-report bias was characteristic of the early years in the study. Significant covariates differed between cotinine and carbon monoxide models. When cotinine was used for verification, about half of the individuals in the smoking intervention group with self-report bias at the first year continued to exhibit bias for 5 years. In absolute terms, the errors associated with measurement were small, but they persisted over 5 years. Some differences appeared to be related to the distinction that carbon monoxide verification was immediate, while cotinine verification was deferred.