In research involving quality of health care, an intervention is often attempted at one or more institutions. This article presents the advantages and disadvantages of various study designs that might be considered for such institution-based research, looking at the issues of comparison groups, replication, blocking, and contamination. Three examples illustrate the importance of accommodating the lack of independence or clustering that occurs among subjects within an institution, and show how these accommodations for clustering can sometimes be made by applying basic statistical techniques to institution-level outcomes, whereas at other times more sophisticated approaches are necessary.