Two criteria are firstly used in the selection of new anticancer agents:--originality of the mechanism of action and significant experimental antitumor activity in an in vivo animal model. Murine tumors grafted in syngenic mice and human tumor xenografts implanted in nude mice are old models which continue to be widely used. Such models are useful but have the tendency to select many false positives (compounds active in mice but inactive in patients). This discrepancy can be explained by differences due to biological materials and also to the methodologies used in laboratories and clinic. Such models will certainly continue to play a major role in the future but they will have to be used in conditions more relevant for clinical extrapolation. Transgenic mice developing in situ tumors constitute new innovative models for in vivo evaluation of anticancer agents. Finally, a putative new class of antitumor agents emerges with the signal transduction modulators: such compounds have antitumor and toxicological properties very different from those of conventional chemotherapeutic agents. If the first representatives prove to be useful in clinic, rules for toxicology, preclinical and clinical evaluation will have to be changed.