Abstract: How to access data transparently is an important issue in information integration research. Grid introduces new challenges to access data transparently, because of the autonomy, diversity and dynamism of data sources. Therefore, higher transparency requirements are enforced in the grid environment, such as name transparency, parallelism transparency and schema change transparency in addition to heterogeneity and location transparency, and they bring stricter requirements to metadata design. A layered architecture of metadata based on the information grid resource space model is introduced to mask the diversity and volatility of physical resources’ schema, to realize independency of physical resource’s name and location, and to realize the goals of transparently accessing resources. Finally, a metadata catalog service that is designed with this metadata architecture, as well as its application in the design and implementation of railway information grid middleware are introduced.