So it seems there is a number of common threads to the comments,
which you should address, since they're likely to surface again.
The most prominent are aperture issues and mixing, typing LINER/HII,
some of the wording, including description of references, and one 
that I feel strongly about, namely quantification of the sorting
effectiveness in various plots.  Draw a curve and estimate
the reliability of classification based on location wrt curve.
Don't be shy about making claims, because that's a primary hook
for others to pay attention to the paper.  Esp for fig 5, but also
updated fig 3 (see below for AFE), and possibly fig 6 (see below).

--I have addressed these issues, including making a more detailed,
  quantitative analysis of Figures 3 and 5, the best new diagnostics.

The focus of the paper needs some sharpening, taking it away
from confirming ISO, toward adding new spitzer-sepcific tools,
hopefully sharper tools.  I really like bringing the low-ionization,
low-excitation lines into the picture, and linking that to the 
OI work -- you should play that up a bit more.  These papers
are a primary source of empirical confirmation of the importance
of XDR lines.  And this could be one of the main themes in this
paper, the other one being in general that the spitzer tools are sharper.

--I have decreased the emphasis on extending ISO work in the abstract.
  And I have added more links to the [OI] work we did with ISO.

The other suggestion to break away from the well-trodden path is
to look at AFE(6.2)/24um rather than th local EQW.  Explore a bit
(using your insights into the SED's), and there is almost certainly
a more discriminating ratio for the Aromatics to be used.

--I have made three relevant plots and posted them at the bottom of
  http://physics.uwyo.edu/~ddale/research/sirtf/lines/DR3/lines.html
  Unfortunately, I did not find that normalizing by 24um provided better
  separation between types.  We are likely experiencing contamination by
  the surrounding PDRs for our HII regions.  i.e., the apertures are large
  enough to make this sensible approach difficult to achieve in practice.

A couple specific things:

1. Fig 6 got some bad press from the paper committee.  
It took me a bit of staring at it
to realize that the real separtion isn't so much the lines you draw
as fits, but the separation along the diagonal.  I can see the -1 slope
diagonal disecting the points into the two groups HII/Sey.  The way
you draw the lines confuses the issue, and led to referees saying 
you overstated the case, which is a reasonable statement in this context.  
In fact, looking at this plot,
I'm starting to see 24um as potentially a key to the discrimination,
not just for normalizing AFE, but also for normalizing these lines.
I think there is still a very powerful diagnostic plot we haven't hit yet,
and I'd guess it involves SiII, SIII, AFE and 24um!

--I have emphasized in the text that the key is the separation along the
  diagonal (not the difference in the slopes).

2. page 6, statement about NED: I think it would be more accurate to say
that the AGN calssification info in NED is incomplete, so that lack of
a flag does not mean lack of AGN.  (As stated now, it sounds like NED
provides generically poor data)

--"incomplete" added

3. The intro sentences to 4.4 sounded a bit trite.  Need to beef up.

--Done