Clinical data repositories represent a potential gold mine of information and knowledge. Rapid access to such information can help bridge the gap between clinical care and research, support clinical and executive decision making, and improve the quality of care. A clinical database can be used in four ways: to display information about an individual patient (results reporting); to find data on a patient with similarities to one being seen (case finding); to describe a group of patients with at least one attribute in common (cohort description); and to analyze data patterns in terms of trends or relationships (predictive modeling). It seems unlikely that many important clinical questions will be subject to randomized clinical trials because of the ethics, logistics, and expense that would be involved. Evolving statistical and epidemiological methods allow us to approach these clinical data repositories with the purpose of building predictive models, but a clear understanding of the limitations of routinely collected clinical data and the inherent biases is necessary. The largest barrier to using routinely collected clinical data is not the limitations of the data themselves, but rather the lack of a data paradigm for the decision-maker. We present some of the problems and pitfalls in obtaining and using routinely collected data, based upon the use of ClinQuery at Boston's Beth Israel Hospital and the resources and traditions at the Mayo Clinic.