Background: Prognostic indexes are useful to guide tailored treatment strategies for cancer patients with brain metastasis (BM). We evaluated the new Graded Prognostic Assessment (GPA) scale in a prospective validation study to compare it with two published prognostic indexes. Methods: A total of 285 newly diagnosed BM (n = 85 with synchronous BM) patients, accrued prospectively between 2000 and 2009, were included in this analysis. Mean age was 62 +/- 12.0 years. The median KPS and number of BM was 70 (range, 20-100) and 3 (range, 1-50), respectively. The majority of primary tumours were lung (53%), or breast (17%) cancers. Treatment was administered to 255 (89.5%) patients. Only a minority of patients could be classified prospectively in a favourable prognostic class: GPA 3.5-4: 3.9%; recursive partitioning analysis (RPA) 1, 8.4% and Basic Score for BM (BSBM) 3, 9.1%. Mean follow-up (FU) time was 5.2 +/- 4.7 months. Results: During the period of FU, 225 (78.9%) patients died. The 6 months-and 1 year-OS was 36.9% and 17.6%, respectively. On multivariate analysis, performance status (P < 0.001), BSBM (P < 0.001), Center (P = 0.007), RPA (P = 0.02) and GPA (P = 0.03) were statistically significant for OS. The survival prediction performances' of all indexes were identical. Noteworthy, the significant OS difference observed within 3 months of diagnosis between the BSBM, RPA and GPA classes/groups was not observed after this cut-off time point. Harrell's concordance indexes C were 0.58, 0.61 and 0.58 for the GPA, BSBM and RPA, respectively. Conclusions: Our data suggest that the new GPA index is a valid prognostic index. In this prospective study, the prediction performance was as good as the BSBM or RPA systems. These published indexes may however have limited long term prognostication capability.