Some factors influencing the detection of malignant cells in sputum samples were evaluated in 449 consecutive cases of primary lung carcinoma seen between 1959 and 1974. Diagnostic accuracy increased during the years under study; the reasons are discussed. The overall accuracy was 82.8%. Detection of malignant cells was 85% for small-cell carcinoma, squamous-cell carcinoma and large-cell carcinoma, 75% for adenocarcinoma, bronchioloalveolar carcinoma and adenosquamous carcinoma and 64% for the uncommon tumors. Accuracy was 87% for central tumors and 42% for peripheral lesions. Tumors less than 2 cm in diameter yielded only 39% accuracy as compared to 90% for larger tumors. The specificity of diagnosis of cell type in those specimens with malignant cells was 95% for small-cell carcinoma and squamous-cell carcinoma, more than 80% for adenocarcinoma and large-cell carcinoma, 65% for bronchioloalveolar-cell carcinoma and adenosquamous carcinoma and less than 30% for the uncommon tumors. Diagnostic accuracy was optimal in those cases with three or more sputum samples: 83% for those with three samples and 90% for those with five or more samples per case. The use of both sputum and bronchial specimens was complementary and increased the accuracy further. Reasons for unsatisfactory specimens included no deep cough, limited cellular material, excessive blood or leukocytes and drying artifacts; the first two were the most common causes.