International Journal of Epidemiology 2002;31:1038–1042

Cigarette smoking is, as a single factor, the most common cause of mortality in developed countries.1 In many of them, declining trends of cigarette consumption have been observed among men, whereas little change or increasing trends have been observed for women.2 It is important to monitor trends in smoking prevalence in order to obtain indications on present and future adverse health effects and to evaluate the impact of activities aimed at its control. It is therefore useful to know the patterns of smoking habits both at national and local level. Many studies have investigated the accuracy of self-reported smoking status using biomarkers as a gold standard (i.e. expired carbon monoxide, salivary thiocyanate, concentration of nicotine and cotinine in plasma, saliva or urine).3–10 These show a con-sistently high validity of self-reported smoking in population-based studies.11 Few studies have been published on non-responder bias in the assessment of smoking habits.12–15 Our study indirectly analyses the effect of non-response bias in a survey on smoking prevalence, comparing the long-term risk of death between responders and non-responders