Published prognostic models for small-cell lung cancer (SCLC) have either combined limited- and extensive-stage patients or have not included standard anatomic staging information to assess the relative value of the knowledge of specific sites and number of sites of metastases in predicting survival in extensive-stage disease. We studied 136 extensive-stage patients in whom traditional staging procedures were performed and in whom other previously demonstrated significant pretreatment variables were determined. Using the Cox proportional hazards model, when all data were included, three variables were significant: performance status (PS) (P = .0001), number of sites of metastases (P = .0010), and age (P = .0029). A prognostic algorithm was developed using these variables, which divided the patients into three distinct groups. When the anatomic staging data were omitted, the serum albumin (P = .0313) was the only variable in addition to PS (P = .0001) and age (P = .0064) that was significant. An alternative algorithm using these three variables was nearly as predictive as the original. Therefore, in extensive-stage patients, reasonable pretreatment prognostic information can be obtained without using the number or specific sites of metastases as variables once the presence of distant metastases has been demonstrated.