Smoking remains a major problem among young people in Europe. However, within the research community examining the issue, debate continues about the best way of assessing the extent of that problem. Questions have been raised about the extent to which existing techniques for generating statistical representations of patterns of youth smoking can address a range of problems connected with identifying, accounting for and correcting unreliable self-report smoking data. Using empirical data from the UK Liverpool Longitudinal Smoking Study (LLSS), this paper argues that self-report measures of smoking, treated in isolation from participants' personal accounts, can disguise problems with the reliability and validity of a given study. Using longitudinal qualitative and quantitative data in dialogue, two main factors contributing to unreliable data are discussed: (a) participants' access to and familiarity with frameworks of everyday cultural knowledge about the practice of smoking, and (b) participants' retrospective revision of events in line with their current goals, aspirations and self-understandings. The conclusion drawn is that research has to employ multiple methods, minimally incorporating some personal contribution from participants, to explore the complex character of the problem of smoking and to avoid the difficulties posed by the models of smoking behaviour embodied within stand-alone statistical research.