To analyze the accuracy and usefulness of sputum cytology as a screening method, 103 cases of histologically proven lung cancer registered from 1998 to 2000 at Kangbuk Samsung Hospital were retrospectively examined. We reviewed the original cytologic and surgical diagnoses for the cases, and the cytology slides of all cytologically negative cases. The overall sensitivity of sputum cytology was 0.83 ; the sensitivity of prebronchoscopy sputum cytology for bronchogenic carcinoma was 0.87. Central tumor location (P=0.002), tumor size (>2.4 cm), (P=0.027) and the number of sputum samples $(geq3)$ (P=0.001) were associated with a positive cytologic diagnosis. Of the 18 cytologically negative cases, 9 cases(38% of smears) were determined to be insufficient for diagnosis, due strictly to low cellularity and saliva. After a review of the cytology slides of cytologically negative cases, we identified several atypical clusters in one case of bronchioloalveolar carcinoma. This negativity was thus attributed to an interpretation error (1/18, 5.6%). Our results suggest that its sensitivity is more strongly related to the specimen adequacy and the times of sampling than to interpretation error. In terms of sensitivity, specificity, accessibility, cost, and morbidity associated with the screening tests, sputum cytology was found to be an accurate effective screening method for lung cancer.