PURPOSE: To report on the outcome of patients with melanoma brain metastases treated with stereotactic radiosurgery (SRS). PATIENTS AND METHODS: One hundred three patients with 153 intracranial melanoma metastases consecutively underwent Linac-based SRS between November 1991 and October 2001. The Kaplan-Meier method, univariate comparisons with log-rank test, and multivariate analyses with classification and regression tree models were performed. Calculations were based on last imaging date rather than the date of the last visit. RESULTS: Median age was 51 years (range, 18-93 years). Median Karnofsky performance status was 90. Sixty-one patients (59%) had single brain metastasis at presentation. Treatment sequence was SRS alone (61 patients), SRS + whole-brain radiotherapy (WBRT) (12 patients), and salvage SRS after WBRT (30 patients). The median tumor volume was 1.9 cm(3) (range, 0.06-22.3 cm(3)). The median SRS minimum peripheral dose and isodose was 18 Gy (range, 10-24 Gy) and 85% (range, 60%-100%), respectively. The median follow-up was 6 months for all patients and 13 months (range, 2-46 months) for patients alive at the time of analysis. The 1-year local control (LC) for all patients treated with SRS was 49%. Among the patients treated with initial SRS alone, the 1-year LC was better for patients with tumors < or =2 cm(3) than with tumors >2 cm(3): 75.2% vs. 42.3% (p < 0.05). The 1-year distant brain metastasis-free survival incidence was 14.7% for the 73 patients receiving either initial SRS alone or SRS +WBRT. The initial number of brain lesions (single vs. multiple) was the only factor with a significant effect on distant brain metastasis-free survival at 1 year: 23.5% for single metastases and 0% for multiple lesions (p < 0.05). The 1-year overall survival was 25.2%. Stratification by Score Index for Radiosurgery (SIR) revealed a significant effect on survival, which was 29% at 1 year for SIR >6 and 10% for SIR <==6 (relative hazard ratio, 2.1; p < 0.05) in classification and regression-tree multivariate analysis involving age, Karnofsky performance status, primary tumor control, tumor volume, SRS dose, SIR (>6 vs. < or =6), and systemic disease status. CONCLUSIONS: Initial SRS alone was an effective treatment modality for smaller cerebral melanoma metastases, achieving a 75% incidence of 1-year LC for < or =2 cm(3) single brain metastases and should be considered in patients with SIR >6. The role of WBRT in melanoma brain metastases cannot be addressed, owing to retrospective bias toward administering this treatment to patients with more aggressive disease. A prospective study is needed to assess the role of WBRT in patients with melanoma brain metastasis.