Despite continuous advances in treatment, cancer remains the leading cause of death in the western world. Current in vitro and in vivo animal models are imperfect predictors of the efficacy of novel cancer drug therapy. Cancer-on-chip technology has shown great promise in improving preclinical models, however validation with clinical trial results and broad adoption has proven challenging due to lack of robustness and scale of the models, and the complexity of human cancer. In this thesis we take a stepwise approach to validating cancer-on-chip models for drug discovery. In this approach cancer-on-chip models are first validated against cancer xenograft models, and subsequently human tumor physiology is added. The specific aim of this thesis is to develop microfluidic cell culture models that can administer dynamic drug concentrations, mimic the drug response as typically found in cancer xenografts, and explore where the chip model can already outperform the xenograft model. Technology was developed to administer dynamic, in vivo-like, drug concentrations on-chip. Efficacy of oxaliplatin against HCT116 colorectal cancer cells was not dependent on peak exposure, as constant exposure and in vivo-like exposure resulted in more growth inhibition than peak exposure only. The administration of in vivo-like oxaliplatin to HCT116 spheroids on-chip resulted in a response similar to what has been found in HCT116 xenografts in mice, both in terms of volume growth and molecular markers of apoptosis and proliferation. It was shown spheroids on-chip were mostly representative of the viable, proliferating shell of the xenograft, which likely determines the drug response. Furthermore a temporary response to oxaliplatin was identified in the chip model which would be hard to detect in xenografts in mice. Besides administering drug concentrations equal to concentrations in the blood of mice, cancer-on-chip systems can be used to test the effect of human drug concentrations. Administration of estimated intracerebral doxorubicin concentrations after Focused Ultrasound, to pediatric brain stem cancer cells (DIPG) cultured in Matrigel on-chip, did not result in growth inhibition. Nevertheless other drugs with favorable ratios of in vitro potency to plasma pharmacokinetics adjusted for blood-brain-barrier passage, were identified and are currently being tested.