The present invention relates to methods for predicting and evaluating metastasis of solid cancers, such as melanoma, in a subject by measuring serum biomarkers associated with a metastatic phenotype. In particular, the present invention provides a serum gene expression signature that is different between highly aggressive and more metastatic versus less aggressive and less metastatic melanomas by quantitatively measuring the levels of, inter alia, lymphoid-specific helicase (HELLS) and condensing complex subunit 2 (NCAPH) transcripts in a subject.