One blinded observer (C.D.S.) retrospectively reviewed 76 previously diagnosed and biopsy-confirmed malignant bronchial brush and wash specimens, 46 non-small cell and 30 small cell carcinomas, obtained from 55 patients. Each case was scored for the presence or absence of 36 standard criteria (architectural, cytoplasmic, and nuclear). Logistic regression analysis was used to determine which criteria were most useful for separating small cell from non-small cell lesions. Although no single criterion displayed 100% sensitivity and specificity for small cell cancer, univariate statistical analysis indicated that 3 individual criteria (nuclear molding, finely granular or "salt and pepper" chromatin, and scant delicate cytoplasm) were more than 90% sensitive and specific in cases of small cell carcinoma. The presence of nuclear molding alone provided the best fit for the logistic regression model. When nuclear molding was present, the odds of a small cell diagnosis increased more than 300-fold. Nuclear molding, finely granular or salt and pepper chromatin, and scant, delicate cytoplasm are the 3 most sensitive and specific cytomorphologic features traditionally used to separate small cell from non-small cell carcinoma. Nuclear molding alone represents the most significant cytomorphologic feature for distinguishing between these malignant lesions.