BACKGROUND: Cancer registries have for decades surveyed the development of cancer diseases. Data on incident cases includes demographic variables. Knowledge of the temporal distribution of risk factors on the same variables makes it possible to model the relationship between disease and risk factor. The results of such analyses might be difficult to interpret since they are based on aggregated data. But the availability of these data sources should encourage further exploration of its possibilities and limitations. METHODS: The temporal pattern of smoking habits in 5-year birth cohorts from 1890-1949 was established, with data on the proportions of current smokers, former smokers and non-smokers and estimated average daily consumption of tobacco and average duration of smoking. The lung cancer incidence among the cohorts in 1953-1992 was analysed by a model which included an additive excess risk for smokers that depended on daily dose and duration of smoking. RESULTS: The lung cancer incidence in later decades was adequately described by the model, which showed a simple relationship with smoking behaviour in the cohorts. For both current smokers and former smokers, the excess risk was about proportional to the daily amount smoked and the 4.5 power of duration of smoking. The age-specific rates for non-smokers were close to a fifth-power curve of age. CONCLUSIONS: Even if lung cancer incidence is not determined separately for groups with known smoking habits, plausible estimates of the effect of smoking can be derived if appropriate information is available on temporal smoking habits in the population.