Some health care organizations allow physicians to withhold cardiopulmonary resuscitation from a patient, despite patient or surrogate requests that it be provided, when they believe it will be more harmful than beneficial. Such cases usually involve patients with terminal diagnoses whose medical teams argue that aggressive treatments are medically inappropriate or likely to be harmful. Although there is state-to-state variability and a considerable judicial gray area about the conditions and mechanisms for refusals to perform CPR, medical teams typically follow a set of clearly defined procedures for these decisions. The procedures are based on the principle of nonmaleficence and typically include consultation with hospital ethics committees, reflecting the guidelines of relevant professional associations. Ethical debates about when CPR can and should be limited tend to rely more on discussions of theory, principles, and case studies than systematic empirical study of the situations in which such limitations are applied. Sociologists of bioethics call for empirical study, arguing that what ethicists and health professionals believe they are doing when they draft policies or invoke principles does not always mirror what is happening on the ground. In this article, we begin the task of modeling the empirical analyses sociologists call for, focusing on a cohort at Massachusetts General Hospital. We inductively analyzed ethics committee notes and medical records of nineteen patients whose surrogates did not accept the decision to withhold CPR.