It is possible that the process of repeatedly measuring the smoking behaviour of adolescents may very well affect that behaviour. This paper reports a test for the extent of such a "Hawthorne" effect in a longitudinal survey of smoking by English adolescents. The self-reported smoking behaviour of 15-16 year olds who attended schools which had participated in the study for five years was compared with that of 15-16 year olds who attended other schools. The prevalence of smoking was lower in those schools which had been surveyed for five years. A number of possible explanations for this finding are discussed. It is concluded that such a "Hawthorne" effect is unlikely to bias analyses relying on comparisons within the data set. However, they can certainly bias the prevalance estimates obtained from such a study. Thus they provide yet another reason why prevalence estimates from cohorts studied over a period of time must be used with considerable caution.