Accurate classification of lung cancer, as well as the differentiation between primary and metastatic carcinoma to the lung, mostly performed on biopsy or fine needle aspiration specimens, is critical for decisions on therapy and for determining prognosis. The limited amount of biopsy material available for morphological assessment has stimulated attempts to improve diagnostic accuracy through the use of immunohistochemistry (IHC), but an optimal IHC diagnostic algorithm has not been firmly established. We evaluated, on a retrospective series of biopsy specimens, the performance of a four-antibody IHC panel for accurate subclassification of non-small cell lung carcinoma (NSCLC) and for identification of metastatic carcinoma. Tumor morphology was assessed and IHC for CK7, CK20, TTF-1, and p63 was performed according to a two-step algorithm. Matched resection specimens served as gold standard and were compared with the corresponding biopsy. Of 443 biopsy specimens studied, 325 were diagnosed as primary carcinoma of the lung, 198 (44.7 %) as adenocarcinoma, 9 (2 %) as possibly adenosquamous carcinoma, 127 (28.7 %) as squamous cell carcinoma, and 40 (9 %) as NSCLC not further classifiable. Ten cases (2.3 %) were classified as adenocarcinoma of unknown origin and 58 (13 %) as metastasis. Importantly, of the primary lung adenocarcinomas, 35 (17.7 %) had been considered on clinical grounds as a metastasis from a previously diagnosed primary tumor. Of the 55 cases submitted to surgical resection in 47 (85.5 %) the biopsy diagnosis was confirmed, revealing substantial agreement (κ value = 0.757). Our two-step approach allows for accurate subclassification of NSCLC and also to distinguish between primary lung adenocarcinoma and metastasis, notably of colorectal adenocarcinoma, with crucial implications for appropriate patient management.