Scientific research relies as much on the dissemination and exchange of data sets as on the publication of conclusions. Accurately tracking the lineage (origin and subsequent processing history) of scientific data sets is thus imperative for the complete documentation of scientific work. However, the lack of a definitive data model for lineage, and the poor fit between current data management tools and scientific software, effectively prevent researchers from determining, preserving, or providing the lineage of the data products they use and create. Based on a comprehensive review of lineage-related research and previous prototype systems, a conceptual framework is presented to help identify and assess basic lineage system components. Within this framework, a direction is outlined for future work on general methods for composing and managing lineage for scientific data.