Despite their common origin from follicular epithelial cells, papillary and follicular thyroid carcinomas differ in their histology and clinical course. In this study the transcriptional profiles of these tumors in comparison with normal thyroid tissue were established. The aim was the development of a molecular tool providing additional information to current histopathological diagnosis and allowing further insight into tumorigenesis. Genome wide expression profiling was performed using Human Unigene Set--RZPD 2 high density cDNA macroarrays comprising 76,000 genes as probes and radioactively labeled cDNA targets retrotranscribed from the isolated RNA of three papillary and three follicular thyroid carcinomas as well as three normal thyroid tissues. 8600 genes differing in their expression between the three groups were selected and printed onto subarrays. Radioactively labeled cDNA targets obtained from 16 papillary carcinomas, 13 follicular carcinomas and 17 normal thyroid tissues were hybridized to these subarrays. 200 genes exhibited a statistically significant expression difference between the two tumor types (p <0.01). In a hierarchical cluster analysis of 124 of these genes (46 known genes and 78 ESTs) the algorythm divided the tumor samples into two groups corresponding to the papillary and follicular thyroid carcinomas. The clearcut diagnostic potential of this method has to be corroborated in a prospective study. Several of the differentiallly expressed genes are known to play a role in tumor development and metastasis. Some of the genes up- or down-regulated in both tumor types are members of known oncogenic pathways in thyroid carcinomas. The complete understanding of complex genome wide expression profiles however awaits a longstanding advancement of hypothesis driven research.