The subjects of the study were 24 patients undergoing biopsies of undiagnosed pulmonary nodules. After a biopsy of a lung nodule was performed and specimens were obtained for pathology, residual cells were placed into a buffer for RNA extraction. Specimens were processed using the modified Eberwine protocol for analysis (U95Av2 array; Affymetrix; Santa Clara, CA), which contains probes for approximately 12,000 genes. We performed a class prediction analysis of the test samples using computer software (GeneSpring, version 5.0; Silicon Genetics; Redwood City, CA), using 300 predictor genes and 3 nearest neighbors. Our training set consisted of 71 microdissected samples, which included non-small cell lung cancer (NSCLC), small cell lung cancer, carcinoid, mesothelioma, normal lung tissue, and pleural tissue. Our gene expression profile prediction categories included NSCLC, small cell lung cancer, and nonneoplastic lung.