The purpose of an enterprise data dictionary is to capture metadata about the data used within the enterprise. The best place to start is with some type of metadata management system (dictionary, repository, etc) that allows you to capture information about your data. This system should allow you to import metadata from numerous sources, including DBMS system tables, schemas, documents, etc. The system should also give you tools that enable you to analyze and compare the metadata you've captured and to document affinities and semantic relationships.
Be aware -- it's a lot of hard work. These steps are similar to the ones required to define master data or to implement a good data integration strategy.
This was first published in August 2008