In this paper, we present a reference model consisting of four different yet equally important perspectives for metadata requirements analysis. We argue that metadata development should start with a systematic requirements analysis as part of a systems engineering approach. This work has been developed as part of the GovStat project, an NSF-funded project designed to explore how US Federal statistical data can be made more understandable and usable by non-expert users. Developing a metadata architecture has been an important part of the project because we believe that applications to support finding and understanding materials from this unique data collection require a properly specified set of metadata. Since we were not able to find an existing metadata schema that meets our needs, we delved into a development of a new schema that was driven by the goals of the project, and by identifying crucial metadata elements through a series of metadata user studies (Denn, Haas, & Hert, 2003). In our effort to evaluate existing standards and define a new metadata schema, we realized that there is a lack of methodological principles and no established framework to inform the process. We believe that such a framework has emerged from our experience (Hert et al., in preparation) and is represented by the four-perspective model in Figure 1. As for methodological principles, we propose to adopt a systems engineering approach. We believe this four-perspective model combined with systems engineering methodologies to be broadly applicable to metadata development initiatives.