Pulmonary neuroendocrine tumors (NE) include a spectrum of tumors from typical carcinoid (TC) to atypical carcinoid (AC), large cell neuroendocrine carcinoma (LCNEC), and small cell carcinoma (SCLC). Little is known about prognostic predictors for AC because of its rarity. Survival analysis was performed on 106 ACs with clinical follow-up from the AFIP and the Pathology Panel of the International Association for the Study of Lung Cancer (IASLC). The tumors fulfilled the 1999 WHO/IASLC criteria for AC of a NE tumor with a mitotic rate of 2 to 10 per 2 mm(2) of viable tumor or coagulative necrosis. Multiple clinical and histologic features were analyzed by Kaplan-Meier and Cox regression analysis. Of the clinical features, higher stage (P = .003) and a tumor size of 3.5 cm or greater (P = .003) were associated with a worse prognosis. Features that were histologically unfavorable by univariate analysis were mitotic rate (P =.002), pleomorphism (P = .018), and aerogenous spread (P =.007). Histologically favorable features by univariate analysis were the presence of palisading (P = .008), papillary (P = .039), pseudoglandular (P =.026), and rosette (P = .022) patterns. Female gender showed a trend toward a poorer prognosis (P =.085) and was included in the multivariate model. Multivariate analysis stratified for stage showed mitoses (P<.001), a tumor size of 3.5 cm or greater (P =.017), and female gender (P =.012) to be the only negative independent predictors of prognosis and the presence of rosettes (P = .016) to be the only independent positive predictor. We further divided the AC into subgroups of low (2 to 5 mitoses/2 mm(2)) and high (6 to 10 mitoses/2 mm(2)) mitotic rate and compared the survival with TC and with LCNEC. Within the category of AC, the patients with a higher mitotic rate had a significantly worse survival than those with a lower mitotic rate (P<.001) stratified for stage. Five- and 10-year survival rates for AC (61% and 35%, respectively) stratified for stage were significantly worse than for TC and better than that for LCNEC and SCLC. Chemotherapy or radiation therapy was given in 12 of 52 and 14 of 52 cases, respectively, but the data were insufficient to evaluate tumor response. We conclude that AC is an aggressive neuroendocrine neoplasm with survival intermediate between TC and LCNEC and SCLC. Higher mitotic rate, tumor size of 3.5 cm or greater, female gender, and presence of rosettes are the only independent predictors of survival. Surgical resection remains the treatment of choice, and the role of chemotherapy and radiation therapy remains to be proven.