The wealth of morphological, histological, and molecular data from human cancers available to pathologists means that pathology is poised to become a truly quantitative systems science. By measuring morphological parameters such as tumor stage and grade, and by measuring molecular biomarkers such as hormone receptor status, pathologists have sometimes accurately predicted what will happen to a patient's tumor. While 'omic' technologies have seemingly improved prognostication and prediction, some molecular 'signatures' are not useful in clinical practice because of the failure to independently validate these approaches. Many associations between gene 'signatures' and clinical response are correlative rather than mechanistic, and such associations are poor predictors of how cellular biochemical networks will behave in perturbed, diseased cells. Using systems biology, the dynamics of reactions in cells and the behavior between cells can be integrated into models of cancer. The challenge is how to integrate multiple data from the clinic into tractable models using mathematical models and systems biology, and how to make the resultant model sufficiently robust to be of practical use. We discuss the difficulties in using mathematics to model cancer, and review some approaches that may be used to allow systems biology to be successfully applied in the clinic.