What I meant was this (that's how I approach duplicate elimination on my empeg that currently holds some 4000 tunes):

- I added 'Refs' and 'FID' column to all views (at least playlist, search and all tracks)
- Select 'All tracks', order by title, look for duplicates or near-duplicates
- Take a look at details to see whether they are duplicates, or, to borrow Tony's example, songs called 'Mother' from several different bands
- When in doubt, listen to them one by one
- If they are duplicates, and one (or more) is not referenced anywhere, delete it; this is the rare case: usually one is referenced in some lists, the other in some more lists
- Look up in CVS list tunes by name and find in which playlists they appear
- Go to those playlists and use FIDs displayed to determine which tune is which (this can be done directly by FID from CVS, but involves hex-to-decimal conversion and back [1])
- Decide which to keep (if they are not identical listening helps), insert it into relevant playlists in stead of the one to be deleted, delete the other

[1] Why didn't I modify spreadsheet containing CVS to display FIDS in hexadecimal? Aha: it wasn't simple to do with list of playlist entries, and I was too lazy to write conversion script, not knowing I have so many duplicates...