The current Atom I-D says to use "application/atom+xml". But it should rather be "text/atom+xml":
The "application/*" MIME type tree is for things, that humans usually won't see, or that are handled only inside applications and thus keep unshown. However the Atom format is an replacement for RSS, and its main purpose is flashing up in feed readers -> it is made for presentation and people, like HTML. Therefore it should get a "text/*" MIME type.
True, there are a few bits in Atom feeds, that aren't usually presented to users (like last-modified timestamps and author names), but that's the case with HTML too (<meta> tags).
Atom is processed by applications, like HTML has specialized applications, but that itself does not mean it should get an application/* tree MIME type. -- MarioSalzer
There is also an IETF mailing list, if someone would like to discuss that point.
I'd say that Atom should use application/*. There are very few cases, I suggest, where using "text" for XML is reasonable; the document would have be small and it's meaning obvious to your average browser user.
FWIW, it's generally understood that "text/html" was a mistake, which is why we went with application/xhtml+xml.
There's also issues with "text/*" types and character encodings that I don't understand, but folks brighter than I do and have recommended against using text/* for this reason. There was even talk of retiring it in the next update to RFC 3023. -- MarkBaker