INTRODUCTION: Published findings on the associations between smoking and the incidence of cutaneous squamous cell carcinoma (SCC) and basal cell carcinoma (BCC) are inconsistent. We aimed to generate prospective evidence on these relationships overall and by anatomical site. METHODS: We followed 1,223,626 women without prior cancer by electronic linkage to national cancer registration data. Questionnaire information about smoking and other factors was recorded at recruitment (1996-2001) and every 3-5 years subsequently. Cox regression yielded adjusted relative risks (RRs) comparing smokers versus never-smokers. RESULTS: After 14 (SD4) years follow-up per woman, 6699 had a first registered cutaneous SCC and 48,666 a first BCC. In current versus never-smokers, SCC incidence was increased (RR = 1.22, 95% CI 1.15-1.31) but BCC incidence was decreased (RR = 0.80, 0.78-0.82). RRs varied substantially by anatomical site; for the limbs, current smoking was associated with an increased incidence of SCC (1.55, 1.41-1.71) and a decreased incidence of BCC (0.72, 0.66-0.79), but for facial lesions there was little association of current smoking with either SCC (0.93, 0.82-1.06) or BCC (0.92, 0.88-0.96). Findings in meta-analyses of results from this and seven other prospective studies were largely dominated by the findings in this study. CONCLUSIONS: Smoking-associated risks for cutaneous SCC and BCC are in the opposite direction to each other and appear to vary by anatomical site.