Multiple copies of a publication reference are, of course, bad news. The Publications system tries to avoid creating duplicates when performing fetch by checking for already loaded DOIs and PMIDs.
However, duplicates may nevertheless be created by mistake, for example when adding a reference manually that in fact has already been added.
A curator may view the Duplicates page which uses a heuristic algorithm to identify possible duplicates.
A curator has to edit and/or delete the duplicated entries by hand to remove the offending entry. Note that labels may have to be transferred from the publication that is to be deleted to the one that is to be kept. This may require admin privileges.