OBJECTIVE To investigate the prognostic factors in non-small cell lung cancer (NSCLC) at stage III and IV and establish a reliable model of clinical prognostic index. METHODS Kaplan-Meier and Cox regression were used to analyze the relationship between the prognostic factors and survival time in 114 cases of NSCLC. The prognostic factors included clinical-pathological features and serum levels of cytokeratin fragment 19 (Cyfra21-1), CEA, neuron-specific enolase (NSE), CA125, interleukin-2 (IL-2) and soluble interleukin-2 receptors (sIL-2R). RESULTS Kaplan-Meier analysis showed that KPS, sex, disease stage, treatment, Cyfra21-1, sIL-2R and CA125 were related to prognosis. Multivariate analysis indicated that Cyfra21-1, stage and treatment were independent prognostic factors. When Cyfra21-1 > 3.5 mg/L, stage IV and chemotherapy < 3 cycles, the relative risk (RR) was 1.691, 2.229 and 3.035, respectively. In patients given 3 or more cycles of chemotherapy, serum Cyfra21-1, sIL-2R and stage at diagnosis were significantly independent prognostic factors. Three of these prognostic factors were used to establish a prognostic index (PI) model based on a simple algorithm: PI = Cyfra21-1 + sIL-2R + stage. The median survival period of patients with 3 or more cycles of chemotherapy were 18 months if PI = 0, 8 months if PI = 1 or 2, and 5 months if PI = 3. CONCLUSIONS The serum Cyfra21-1, sIL-2R and disease stage in unresectable NSCLC were independent prognostic factors. PI calculated on the basis of Cyfra21-1, sIL-2R and stage is recommended to predict the survival period of NSCLC.