When importing references, or when using Tools -> Find duplicates, JabRef displays side-by-side potential duplicated references. It is then up to you to keep one of them, or to merge these references.
To help you in this task, JabRef currently offer 5 comparison modes.
Which one do you use?
More important to me is that my latest version Win7 instal from what I can work out JabRef checks on year,journal,author (not bibtexkey) and if any of these are different it doesn’t detect the duplicate!!! This is a real issue given that many web sites provide the BibLaTeX date and journaltitle fields meaning they are NEVER detected by JabRey.
Also Web importing from Google into JabRef is much more consistent at detecting duplicates than direct Searching on author year title in JabRef! IMHO the search/duplicate functionality requires improving and making able to deal with typos (cut and paste from OCR pdfs) like Google appears able to with exactly the same author year title ie: a fuzzy search option.
Where are the comparison modes set? My impression is that the current comparison version is hardcoded as discussed in More control on the Duplicate Finder