Much has been written about the advantages and disadvantages of various oncology model systems, with the overall finding that these models lack the predictive power required to translate preclinical efficacy into clinical activity. Despite assertions that some preclinical model systems are superior to others, no single model can suffice to inform preclinical target validation and molecule selection. This perspective provides a balanced albeit critical view of these claims of superiority and outlines a framework for the proper use of existing preclinical models for drug testing and discovery. We also highlight gaps in oncology mouse models and discuss general and pervasive model-independent shortcomings in preclinical oncology work, and we propose ways to address these issues.