Esophageal squamous cell carcinoma has heterogeneous clinical outcomes that cannot be predicted well using any existing clinical or molecular prognostic factors. Gene expression profiling may enable more precise prediction of the clinical outcome of these patients. We developed a new approach using gene expression ratios of paired cancerous and normal tissue specimens from the same patient to reduce the effects of variation among individuals. Using oligonucleotide microarrays, we analyzed total RNA expression levels corresponding to 12,600 transcript sequences in 24 paired cancerous and normal tissue operative specimens from 12 patients with esophageal squamous cell carcinoma. Hierarchical clustering analysis using gene expression ratios (cancer:normal) divided the 12 patients into two groups; all 7 patients in the first cluster survived without relapse (median follow-up, 483 days), whereas all 5 patients in the second cluster relapsed (median relapse-free survival time, 280 days; log-rank test, P = 0.006). In contrast, clustering either with cancerous tissue alone or with normal tissue alone did not show significant differences in the outcomes. The expressions of a variety of genes related with cell cycle, gene-repair, apoptosis and chemoradiotherpay resistance were up-regulated in the poor prognostic cluster. These results suggest that ratios of paired gene expression profiles may more efficiently predict relapse-free survival of esophageal squamous cell carcinoma than existing prognostic factors or than gene expression profiling with cancerous tissue alone.