The use of archival sources is understood to be an important research tool, but the problems involved have rarely been distinguished from the broader difficulties of interpreting qualitative sources. Attempts to use archival material for hypothesis testing, as opposed to description or theory development, are confounded by the large size and often opaque structure of archives; factors which lead to misinterpretations of evidence and a tendency to confirm the author's expectations. This paper discusses common features of archival materials, and shows how they can compound traditional research design problems. It then proposes a set of best practices for avoiding these problems, most notably the use of strong and explicit sampling procedures. These practices are illustrated using a brief discussion of material from the National Archives of India on the 1975 Emergency.