Electronic health databases promise to transform both the assessment of health care delivery and our understanding of treatments' safety and effectiveness. To achieve these goals, it will be necessary to (1) recognize limits on inferring causality; (2) protect confidentiality while allowing important societal gain; (3) link health data back to the individual patient; (4) obtain additional information from medical records; (5) understand ways in which electronic data can misrepresent reality; and (6) create the infrastructure, expertise, and resources to use the data. Realizing databases' potential will require long-term commitment and investment beyond the maintenance of the databases themselves.