Squamous cell carcinomas of the head and neck constitute an anatomically heterogeneous group of neoplasms that share in common a causal association with tobacco and alcohol exposure. The clinical course of these neoplasms is difficult to predict based on established prognostic clinicopathological criteria. Given the genetic complexity of head and neck cancers, it is not surprising that correlations with individual genetic abnormalities have also been disappointing. Several authors have suggested that global gene expression patterns can be used to subgroup patients with cancer. Here we report the use of cDNA microarrays containing 9216 clones to measure global patterns of gene expression in these neoplasms. We have used a statistical algorithm to identify 375 genes, which divide patients with head and neck tumors into two clinically distinct subgroups based on gene expression patterns. Our results demonstrate that gene expression profiling can be used as a predictor of outcome.