Effective development of targeted anticancer agents includes the definition of the optimal biological dose and biomarkers of drug activity. Currently available preclinical models are not optimal to this end. We aimed at generating a model for translational drug development using pancreatic cancer as a prototype. Resected pancreatic cancers from 14 patients were xenografted and expanded in successive groups of nude mice to develop cohorts of tumor-bearing mice suitable for drug therapy in simulated early clinical trials. The xenografted tumors maintain their fundamental genotypic features despite serial passages and recapitulate the genetic heterogeneity of pancreatic cancer. The in vivo platform is useful for integrating drug screening with biomarker discovery. Passages of tumors in successive cohorts of mice do not change their susceptibility to anticancer agents and represent a perpetual live bank, facilitating the application of new technologies that will result in the creation of an integrated stable database of tumor-drug response data and biomarkers.