The prognosis of men with moderate-grade prostate cancer is uncertain. At present, there are few if any reliable molecular markers that can distinguish moderate-grade tumors from those that behave more aggressively. To better understand the molecular basis of human prostate cancer and potentially provide information toward more accurate prognosis, we measured and analyzed gene expression profiles of 13 high- and moderate-grade human prostate tumors using cDNA microarrays. The expression of 136 genes was observed to differ significantly (P < 0.001) between normal prostate and tumors using one-sample t testing and Wilcoxon ranking. Hierarchical clustering of genes demonstrated a relatively similar pattern of differential expression across the tumors. However, importantly, permutation t tests (two-tailed P < 0.001) revealed 21 genes whose expression profiles segregated moderate- and high-grade tumors from each other, which was significantly (P < 0.03) greater than what was expected by chance. These results were compared in silico with prostate cancer profiling efforts performed by other groups, including a meta-analysis of four data sets, which validated many of the dysregulated genes. We suggest that these data provide insight into the molecular nature of clinically aggressive prostate cancer.