Well, it doesn't have to therefore be a book using Swing exactly,
however there is no way to broach this topic *without* getting into
philosophy/psychology of user interface designs.

Best so far is About Face 2.0: The Essentials of Interaction Design.

But really, all its a matter of doing is making things look like other
things you've already seen -- take advantage of research already done
in the form of successful products already out there.  Don't try a new
layout or interaction metephore without 1) ruling out the existing
ones as not being able to handle the necessary interaction, 2) trying
it out on test users before calling it "final", and 3) really being
sure that the interaction you've designed is absolutely necessary (can
it be simplified to metaphores already known).

It all comes down to "expected behavior".  What would a user expect to
see, based on the applications they already know (assume they know
Office, IE, Wordpad, a standard mailer like Outlook, Outlook Express,
or Eudora)?

Then realize there are places in Swing where the expected isn't and
you'll need to do derivations and overrides to get it that way.  E.g.,
menu items shouldn't have tooltips, but creating a JMenuItem with an
Action will create one that you'll have to undo; toolbar icons
shouldn't have text and should all be the same size, but creating one
with an Action will set the text and the button sizes may vary if your
icons are different sizes; hitting a key in a non-editable JComboBox
should skip to the item starting with that letter, but that works on
the String.valueOf of the model item, not what the renderer may be
showing.

Layout Management is key.  Learn GridBag, learn Border, maybe try the
new layouts in Swing for 1.4.  I haven't had much luck with them, but
then again, I've been good at GridBag since it started as the XmpTable
widget for X-Windows/Motif.

Its lots of little things, and there's no one catalog of all of them.
Its all a matter of experience and comparisons.  If it doesn't act the
same way a typical windows app might act, there had better be a reason
for it, or go fix it.  This is extremely time-consuming.  Anybody can
get a quick and dirty swing, gnome, or windows app out there, but
it'll take *weeks* of time actually giving it the polish that a
professional work might have, with a lot of nitty-gritty digging into
the implementation of components and googling for examples to figure
out what the hell to actually do.  Be sure your customer base (or at
least your boss) is willing to pay for that, but don't think you won't
have to sell the idea.

That's about all I can say on it.

Joe