Uveal melanomas frequently metastasize and cause patient death. Many clinical, histopathologic, molecular, and genetic factors have been linked to metastasis. We hypothesized that understanding the relationships between, and relative prognostic significance of these factors would provide new insights into the pathogenesis of metastasis. To this end, we collected clinical, pathologic, and molecular data for 65 uveal melanomas, including patient age, sex, tumor size, location, cell type, vasculogenic mimicry looping matrix patterns, gene expression profiles, and immunohistochemistry for cytokeratin-18, vascular endothelial cadherin, E-cadherin, beta-catenin, and hypoxia-inducible factor 1alpha. In addition, we used Gene Set Enrichment Analysis to identify statistically significant overlap in genes that were differentially expressed in metastasizing tumors and those expressed in other well-characterized biological systems. Our results show that the class 2 gene expression signature was the most accurate predictor of metastasis (P=0.0001) and that the biomarkers most strongly associated with the class 2 signature included epithelioid cell type, beta-catenin, E-cadherin, and hypoxia-inducible factor 1alpha (P< or =0.001 for each). Thus, the class 2 gene expression signature continues to be the most accurate predictor of uveal melanoma metastasis and can, therefore, serve as a benchmark for evaluating other biomarkers. Importantly, Gene Set Enrichment Analysis showed a significant association between genes expressed in class 2 tumors and those expressed in primitive ectodermal and neural stem cells. Taken together with the constellation of biomarkers associated with the class 2 signature, this suggests the presence of cancer cells with a primitive neural/ectodermal stem cell-like phenotype that may be responsible for metastasis in these highly aggressive tumors.