BACKGROUND There is a need for early detection methods for lung cancer. Radiologic imaging may be more sensitive for peripheral cancers than for cancers arising in the central airways, from which bronchial epithelial cells are exfoliated into the sputum. METHODS Sputum samples were collected at baseline and periodically thereafter in a cohort of smokers and former smokers with chronic obstructive lung disease. The association between cytologic atypia and incident lung cancer was assessed by hazard ratios (HR; 95% confidence intervals) using Cox regression and by odds ratios (95% confidence intervals) using logistic regression, adjusting for potential confounding factors. RESULTS We observed 174 incident lung cancers in a cohort of 2,521 people over 9,869 person-years of observation. Risk for incident lung cancer was increased among those with cytologic atypia graded as moderate or worse (adjusted HR, 2.37; 1.68-3.34). The association between sputum atypia and lung cancer incidence was greatest for those sputum samples collected 5 months or less before the diagnosis of lung cancer (odds ratio, 10.32; 5.34-19.97). The association was substantially stronger for squamous cell lung cancers (HR, 5.13; 2.89-9.10) than for adenocarcinomas (HR, 1.85; 0.94-3.65). CONCLUSIONS Cytologic atypia is a marker for increased lung cancer risk. These cytologic changes seem to arise from late events that are most apparent for cancers arising in the central respiratory airways. Whether cytologic atypia might complement radiologic imaging in a combined approach to lung cancer, early detection requires additional evaluation of those two methods used together.