Prostate cancer is the most commonly diagnosed cancer in men. In the August 23 Nature, Saravana Dhanasekaran and colleagues from the University of Michigan Medical School describe the use of cDNA microarrays to define 'signature' gene-expression profiles for human prostate cancer (Nature 2001, 412:822-826). They compared the expression levels of almost ten thousand genes in benign and malignant prostate cancer samples, compared with normal adjacent prostate tissue. Clustering analysis revealed differences between normal and malignant samples, and between metastatic and localized prostate tumours. Dhanasekaran et al. used tissue microarrays to confirm the upregulation in neoplastic samples of the cell-surface serine protease hepsin and the proto-oncogenge kinase PIM1. Thus, the combination of DNA microarrays and tissue microarrays with clinical data offers a powerful strategy to define prognostic markers for prostate cancer.