Historically, cancer drug development has been a roller coaster. Numerous agents have shown exciting activity in preclinical models and yet have had minimal activity clinically. These disappointments have led to reasonable scepticism about the true value of both syngeneic and xenograft rodent tumour models in accurately identifying agents that will have important clinical utility. Whereas the development of newer techniques, including transgenic mouse models of cancer, offers the potential to develop more predictive models, the role of such mice in cancer drug development is not yet validated. To advance in our understanding of predictive model systems it may be wise to analyse both the successes and the failures of conventional models in order to understand some of their limitations and perhaps to avoid making the same mistakes in the future. Here we review the value and limitations of xenograft models, and the role of integrating preclinical pharmacology in developing new treatments for solid tumours of childhood.