Deficiencies in the standard preclinical methods for evaluating potential anticancer drugs,such as xenograft mouse models, have been highlighted as a key obstacle in the translation of the major advances in basic cancer research into meaningful clinical benefits. In this article, we discuss the established uses and limitations of xenograft mouse models for cancer drug development, and then describe the opportunities and challenges in the application of novel genetically engineered mouse models that more faithfully mimic the genetic and biological evolution of human cancers. Greater use of such models in target validation, assessment of tumour response, investigation of pharmacodynamic markers of drug action, modelling resistance and understanding toxicity has the potential to markedly improve the success of cancer drug development.