cDNAmicroarrays, combined with bioinformatics analyses, are becomingincreasingly used in current medical research. Existing analytic methods,particularly those that are unsupervised, often have difficulty recognizing subtle differences among predefined subgroups. In contrast, supervised methods, such as Artificial Neural Networks (ANNs), are able to recognize subtly different biological entities. We applied ANNs in a proof-of-principle study of cDNA microarray data in esophageal cancer (CA) and premalignancy. cDNA microarrays, each containing 8064 clones, were hybridized to RNAs from 22 esophageal lesions, including 14 Barrett's esophagus (BA) metaplasias and 8 esophageal carcinomas (3 squamous cell carcinomas and 5 adenocarcinomas). Scanned cDNA microarray data were analyzed using the bioinformatics software Cluster/TreeView, Significance Analysis of Microarrays (SAM), and ANNs. Cluster analysis based on all 8064 clones on the microarrays was unable to correctly distinguish BA specimens from CA specimens. SAM then selected 160 differentially expressed genes between Barrett's and cancer. Cluster analysis based on this reduced set still misclassified 2 Barrett's as cancers. The ANN was trained on 12 samples and tested against the remaining 10 samples. Using the 160 selected genes, the ANN correctly diagnosed all 10 samples in the test set. Finally, the 160 genes selected by SAM may merit further study as biomarkers of neoplastic progression in the esophagus, as well as in elucidating pathological mechanisms underlying BA and CA.