PURPOSE: This study evaluates the prognostic performance of a 15 gene expression profiling (GEP) assay that assigns primary posterior uveal melanomas to prognostic subgroups: class 1 (low metastatic risk) and class 2 (high metastatic risk). DESIGN: Prospective, multicenter study. PARTICIPANTS: A total of 459 patients with posterior uveal melanoma were enrolled from 12 independent centers. TESTING: Tumors were classified by GEP as class 1 or class 2. The first 260 samples were also analyzed for chromosome 3 status using a single nucleotide polymorphism assay. Net reclassification improvement analysis was performed to compare the prognostic accuracy of GEP with the 7th edition clinical Tumor-Node-Metastasis (TNM) classification and chromosome 3 status. MAIN OUTCOME MEASURES: Patients were managed for their primary tumor and monitored for metastasis. RESULTS: The GEP assay successfully classified 446 of 459 cases (97.2%). The GEP was class 1 in 276 cases (61.9%) and class 2 in 170 cases (38.1%). Median follow-up was 17.4 months (mean, 18.0 months). Metastasis was detected in 3 class 1 cases (1.1%) and 44 class 2 cases (25.9%) (log-rank test, P<10(-14)). Although there was an association between GEP class 2 and monosomy 3 (Fisher exact test, P<0.0001), 54 of 260 tumors (20.8%) were discordant for GEP and chromosome 3 status, among which GEP demonstrated superior prognostic accuracy (log-rank test, P = 0.0001). By using multivariate Cox modeling, GEP class had a stronger independent association with metastasis than any other prognostic factor (P<0.0001). Chromosome 3 status did not contribute additional prognostic information that was independent of GEP (P = 0.2). At 3 years follow-up, the net reclassification improvement of GEP over TNM classification was 0.43 (P = 0.001) and 0.38 (P = 0.004) over chromosome 3 status. CONCLUSIONS: The GEP assay had a high technical success rate and was the most accurate prognostic marker among all of the factors analyzed. The GEP provided a highly significant improvement in prognostic accuracy over clinical TNM classification and chromosome 3 status. Chromosome 3 status did not provide prognostic information that was independent of GEP.