Mouse models of cancer have consistently been used to qualify new anticancer drugs for study in human clinical trials. The most used models include transplantable murine tumors grown in syngeneic hosts and xenografts of human tumors grown in immunodeficient mice. For the latter systems, retrospective preclinical-clinical correlation studies are available, which suggest that improvements must be made to increase their value. Transgenic, knock-out, and knock-in mouse models and their intercrosses are more recent developments that mirror defined steps of human carcinogenesis. However, their value in predicting clinical results remains to date poorly defined. We take the position that properly used and interpreted human tumor xenografts grown in immunodeficient mice can be useful, although not absolutely predictive of behavior in the clinic, and continue to make contributions to critical clinical development choices. (cancer Res 2006;66(7):3351-4)