Neuroendocrine tumors of the lung embrace a spectrum from low-grade typical carcinoid (TC), intermediate-grade atypical carcinoid (AC), and high-grade categories of large cell neuroendocrine carcinoma (LCNEC) and small cell carcinoma (SCLC). We studied 200 neuroendocrine lung tumors to critically evaluate the Arrigoni histologic criteria for AC using statistical analysis to delimit more rigorously an intermediate survival for AC between TC and the high-grade tumors of LCNEC and SCLC. Histologic features that might predict prognosis were used for Kaplan-Meier and Cox proportional hazards survival analysis, and an optimal mitotic range for AC was calculated. The optimal mitotic range for AC was 2 to 10 mitoses per 2 mm2 of viable tumor (10 high-power fields). Based on this finding, we collapsed mitoses into three categories (< 2; 2-10; > or = 11) and performed Cox multivariate analysis for all 200 neuroendocrine tumors. Mitotic counts were the only independent predictor of prognosis. Based on this analysis, we propose that AC be defined as a tumor with neuroendocrine morphology, mitotic counts between 2-10 per 2 mm2 of viable tumor (10 high-power fields), or coagulative necrosis. Using these criteria, the 200 neuroendocrine tumors were classified as 51 TC, 62 AC, 37 LCNEC, and 50 SCLC. The 5- and 10-year survival was 87% and 87% for TC, 56% and 35% for AC, 27% and 9% for LCNEC, and 9% and 5% for SCLC, respectively. After stratification for stage, survival for AC was significantly worse than for TC (p < 0.001); for LCNEC and SCLC it was significantly worse than for AC; but the survival for LCNEC was no different than that for SCLC.