Descriptions are created as word processor documents in Microsoft Word, using a template which structures the description according to the elements of ISAD(G). The document creator uses dialogues provided by Word macros to provide additional structuring of the text at the sub-paragraph level. (Where non-digital descriptions existed, they have been re-keyed). The SGML markup is provided by a Word macro which processes the (highly structured) Word document produced by using the template described in the previous paragraph. The Word conversion macro maps the ISAD(G) elements to the appropriate EAD element types, generates standard EAD header information, and outputs a complete valid EAD-encoded document. (N.B. the macro is *not* intended as a generic Word-to-SGML tool: it is designed specifically to process the structured document which is produced from using the template in a controlled manner.) Some limited checking of structure and content is performed by Word macros either at the time of document creation or at the point of conversion to SGML. That SGML document is then processed and validated by James Clark's NSGMLS parser. The SGML document is converted to an XML version using James Clark's SX, and that provides the input for the XSL(T) processing described above to generate HTML renditions. The document creator uses a thesaurus lookup procedure for all access point terms to ensure that occurrences of such terms (a) have a standard form and content, and (b) are associated with a unique identifier for the entity which acts as a pointer to an archival authority record for that entity. At the time of writing (October 1999), however, we do not have a search and retrieval tool which can fully exploit that markup. Some very basic static index pages for the descriptions are currently generated through the use of XSL(T) stylesheets. All the HTML renditions of our descriptions contain (as HTML meta elements) basic Dublin Core metadata derived from the content of the EAD header and archdesc-level controlaccess elements. We are also experimenting with the generation of RDF-based metadata for the finding aids. At present, this employs the semantics of Dublin Core, but we envisage that it could be extended to incorporate other metadata schemas as required.