Tumor classification has evolved based upon tumor site, mostly an organ but occasionally a system, such as the hematopoietic or lymphoid system, and on tumor histological characteristics. Using this relatively straightforward approach, a globally accepted taxonomy has been developed, as reflected in the WHO classification of human tumors. With the development of methods to explore the molecular composition of a cell or tissue sample in situ, including immunohistochemistry for proteins and in situ hybridisation for nucleic acids, a new level of complexity developed in tumor classification. This has culminated, in our post-genome era, in detailed molecular mapping of tumor tissue, even at the level of the individual cell, or of tumor DNA or RNA in circulating blood. Such approaches are having a profound impact on tumor classification, to the extent that now entities emerge that can only be diagnosed by molecular analysis. In this article we will review in some detail this development and discuss how molecular analysis might be integrated into diagnostic pathology, in order to get to a comprehensive approach to tumor classification.