One of the most serious obstacles facing investigators involved in the development and assessment of new anti-cancer drugs is the failure of preclinical rodent tumor models to predict in a reliable way whether a given drug will have anti-tumor activity and acceptable toxicity in humans. Most previous investigations for assessing drug activity in vivo have utilized rapidly growing non-metastatic transplantable mouse or human tumors injected ectopically in syngeneic or nude mice, respectively. Some of the reasons for the inadequacy of such models are well known and, as a result, there has been a gradual movement toward the use of transgenic oncomouse models for anti-cancer drug testing. It is too early to conclude, one way or the other, whether these will be superior to transplantable tumor models. Moreover, such transgenic models have a number of limitations which are not widely appreciated. It is argued that transplantable tumor models, with various modifications, might be made significantly more predictive than current models, and would thus constitute a more economic alternative to the use of large numbers of transgenic oncomice. These modifications include the use of slower growing and genetically tagged (e.g. LacZ or GFP) tumors which are transplanted initially into orthotopic organ sites. These methods would facilitate the growth and detection of distant microscopic and macroscopic metastases, the response of which to anti-cancer drugs, using 'clinically equivalent doses,' could be evaluated.