This paper discusses the conceptual challenges faced in designing a new system of matching incoming records for a very large database from diverse sources. Problems of satisfying a “match” with sufficient flexibility and rigor in an environment of imperfect data are outlined.