There are two ways you can approach this.

If there is some simple field or combination of fields that you can use to
determine which letter needs to be produced, then you can try the following:
a. Let's suppose that the field "Letter" contains code A,B,C,D,E,F
a. set up the 6 letters, with the Merge fields you need. Put them in
c:\letters\a.doc, c:\letters\b.doc, c:\letters\c.doc, etc.
b. set up a new letter and attach it to your data source in the usual way.
Then use an { INCLUDETEXT } field to include the correct letter. It will
need to be something like

{ INCLUDETEXT "c:\\letters\\{ MERGEFIELD Letter }.doc" }

Then attempt to preview the merge.

If each letter has different headers and footers, you may find that you have
to bookmark the header, body (leave out the final paragraph mark), and
footer using names such as "header", "body" and "footer". Then instead of
the above, use

{ INCLUDETEXT "c:\\letters\\{ MERGEFIELD Letter }.doc" body }

and insert equivalent INCLUDETEXT fields in the header and footer of the
mail merge main document.

However, the more complication you introduce, the less likely this approach
is to work, and I suspect it is less reliable in Word 2002 than Word 2000.
Also, if there is no simple field or fields you can use to map to a document
name for each record, you can't use the approach. So another method would be
to use VBA to open the data source and the 6 letters and use automation to
execute one merge for each record in the data source. Precisely how to do
this would depend on the data source and how familiar you are with VBA.

--
Peter Jamieson
MS Word MVP