The success rate in predicting clinical efficacy of anti-cancer modalities using current xenograft models has been reported to be only 30-40%. Standard xenograft models use cell lines that are maintained in plastic and have adapted to grow independently of the tumour microenvironment, resulting in models with genetic and phenotypic characteristics distinct from that seen in the clinic. Rajendra Kumari and Martin Page at PRECOS look at why, in the attempt to reduce drug attrition and improve clinical predictivity, patient-derived xenograft tumours are being used to improve and refine preclinical modelling.