Awareness of the role in tumor biology of epigenetics and transcription regulatory mechanisms, such as miRNAs, is increasing. We assume that transcription factors play a crucial role in that context, such that they might be capable of discriminating different tumor classes and yet unknown subgroups. Our approach aims at unsupervised identification of a) sample subsets within a dataset and b) the TFs associated with these subsets. The results show, that an unsupervised selection of TFs can clearly distinguish between the classic breast cancer subtypes and different tumor types such as breast, colon, kidney or prostata cancer. Therefore we assume that the method might help in future to identify new markers suitable for diagnosis of specific treatment related phenotypes or detection of tissues of origin.