Microarray gene expression profiling is a powerful tool for generating molecular cancer classifications. However, elucidating biological insights from these large data sets has been challenging. Previously, we identified a gene expression-based classification of primary uveal melanomas that accurately predicts metastatic death. Class 1 tumors have a low risk and class 2 tumors a high risk for metastatic death. Here, we used genes that discriminate these tumor classes to identify biological correlates of the aggressive class 2 signature. A search for Gene Ontology categories enriched in our class-discriminating gene list revealed a global down-regulation of neural crest and melanocyte-specific genes and an up-regulation of epithelial genes in class 2 tumors. Correspondingly, class 2 tumors exhibited epithelial features, such as polygonal cell morphology, up-regulation of the epithelial adhesion molecule E-cadherin, colocalization of E-cadherin and β-catenin to the plasma membrane, and formation of cell-cell adhesions and acinar structures. One of our top class-discriminating genes was the helix-loop-helix inhibitor ID2, which was strongly down-regulated in class 2 tumors. The class 2 phenotype could be recapitulated by eliminating Id2 in cultured class 1 human uveal melanoma cells and in a mouse ocular melanoma model. Id2 seemed to suppress the epithelial-like class 2 phenotype by inhibiting an activator of the E-cadherin promoter. Consequently, Id2 loss triggered up-regulation of E-cadherin, which in turn promoted anchorage-independent cell growth, a likely antecedent to metastasis. These findings reveal new roles for Id2 and E-cadherin in uveal melanoma progression, and they identify potential targets for therapeutic intervention.