RATIONALE Global gene expression analysis provides a comprehensive molecular characterization of non-small cell lung cancer (NSCLC). OBJECTIVES To evaluate the feasibility of integrating expression profiling into routine clinical work-up by including both surgical as well as minute bronchoscopic biopsies and to develop a robust prognostic gene expression signature. METHODS Tissue samples from 41 chemotherapy-naïve non-small cell lung cancer patients and 15 control patients with inflammatory lung diseases were obtained during routine clinical work-up and gene expression profiles were gained using a oligonucleotide array platform (NovaChip; 34'207 transcripts). Gene expression signatures were analyzed for correlation with histological and clinical parameters and validated on independent published datasets and immunohistochemistry. MEASUREMENTS AND RESULTS Diagnostic signatures for adenocarcinoma and squamous-cell carcinoma reached a sensitivity of 80%/80% and a specificity of 83%/94%, respectively, dependent on the proportion of tumor cells. Sixty-seven of the 100 most discriminating genes were validated with independent observations from the literature. A 13-gene metagene refined on 4 external datasets was built and validated on an independent dataset. The metagene was a strong predictor of survival in our dataset (HR=7.7, 95% CI [2.8-21.2]) and in the independent dataset (HR=1.6, 95% CI [1.2-2.2]) and in both cases independent of the UICC-staging. Vascular endothelial growth factor-beta, one of the key prognostic genes, was further validated by immunohistochemistry on 508 independent tumor samples. CONCLUSIONS Integration of functional genomics from small bronchoscopic biopsies allows molecular tumor classification and prediction of survival in NSCLC and might become a powerful adjunct for the daily clinical practice.