As digital library development begins to focus on interoperability and collection federation, resource developers need to be concerned with contributing to national and international collections, while not losing sight of the needs of institutions and user communities. The Digital Collections and Content (DCC) project aims to provide integrated access to IMLS National Leadership Grant (NLG) digital collections through a centralized collection registry and metadata repository. While technical development proceeds on the repository, our research team is investigating how collections and items can best be represented to meet the needs of both service providers and diverse user communities. This paper presents results on metadata and collection representation practices based on survey data, interviews, and content analysis. Despite Dublin Core's prevalence and ease of use, problems with field richness and consistency of application persist, in part because of the distinct cultures of description that have evolved in different kinds of cultural heritage institutions. Moreover, the concept of a digital collection is widely unsettled among resource developers. This has important implications for central repositories, if, as we hypothesize, the strategic foregrounding and backgrounding of collection-level metadata proves critical for navigation and interpretation of information in large-scale federated collections.