The widely used multistage model of Armitage and Doll is fit to the British physician lung cancer data of Doll and Hill under the assumption that cigarette smoke induces the initial and penultimate changes. It is shown that the best fit of this model in continuing smokers gives predictions not in accordance with incidence in ex-smokers and dose-response. A better global fit can be obtained by increasing the number of stages, but this de-emphasizes initiation and is inconsistent with the rise of incidence in nonsmokers. Thus, one should look to other models. A two-stage model with clonal growth in which smoking initiates normal target cells and promotes the clonal growth of just the smoke-initiated cells is proposed. This model is shown to agree with the Doll and Hill data and thus it has empirical plausibility that should encourage biological studies of clonal growth in carcinogenesis.