AbstractFusion experiments produce large amounts of raw data which must be evaluated. From these data the diagnosticians derive ‘processed’ data files in increasing ‘levels’ of evaluation. The processed data files can be modified several times due, for example, to recalibration or correction of errors. The re-computation of files and the invalidation of the previous versions cause consistency problems. The dependence data base (DDB) addresses this problem, which is essential for the users of experimental data. The DDB keeps track of the mutual dependences between data files. Any re-computation creates a new ‘valid’ version of a file and renders the previous version and all files derived from it ‘invalid’. Previous versions are kept in the file system and a file history can be retrieved. The DDB works on-line, so that it is always up-to-date. The structure of the DDB and its application to ASDEX-Upgrade data are described.