sec 3.3 Aside from the PAH 6.2um EW, only line fluxes are needed for the diagnostics, and the distinction between "sky" continuum and dust continuum is immaterial. I therefore didn't understand the discussion of why it was important to use MIPS 24um to subtract the sky emission. --In the (pre-)development stages of this paper, I made a large number of exploratory plots, and some included line-to-continuum ratios. Section 3.3 aims to clarify that we can't use formal equivalent widths for Long-High data (since we didn't take sky data to properly remove the sy continuum). As a useful substitute for the Long-High continuum we use 24um data. If more co-authors are likewise confused, I will re-work this section. sec 4.2 The text states that "PAH features in the mid-IR have been used to characterize the physical state of the gas in PDRs". What exactly does this refer to? --I don't know.(!) I see that back in Version 2.1 I had "physical state of a system" - I basically meant to imply that folks like Genzel have used them to characterize if a system is AGN-dominated versus powered by star formation. I'm unsure why I changed the text to its current wording. I have changed the text to better reflect my original intent. The text claims that ...the blue and red sides of the continuum are easy to define. I suggest putting "continuum" in quotes -- as is made clear in the paper JD has written on the PAH features, the 6.2um feature sits on top of the red wing of the 7.7um complex. About 40% of the 6.2um power is missed by the "spline continuum" placement. --Done sec 5.2.1 Is PAH EW (i.e., ratio of PAH/continuum) the best thing to use for classification? This 6.2um continuum could be from hot dust, but could also be from starlight or AGN continuum. I appreciate that PAH EW has the advantage of being "local" (i.e., only using information spectrally close to 6.2um) but it might not be the best diagnostic of the dust properties. If the information were available, I think that L(PAH)/L_TIR would be more "diagnostic" of the properties of the dust. But I appreciate that L_TIR requires MIPS, and the 160um psf is not well-matched to the 23"x15" spectral extraction apertures. So, as a compromise, since you have the 5.5-38um spectrum, it might be interesting to try plotting [OIV]/[NeII] vs L(PAH6.2)/L(6-38um), and [SiII]/[NeII] vs L(PAH6.2)/L(6-38um). I wonder how these would compare with the current Fig. 3 in terms of separating star-powered vs AGN-powered emission regions? --I was aware of this issue, and have toyed with different types of normalization. What makes your specific suggestion difficult is that we don't have 6-38um fluxes for extranuclear or archival targets, which comprise over half of the sources in these figures. Using 24um is also problematic since it's quite sensitive to the far-IR color; using 24um as the normalizer leads to a much less coherent diagram. Somewhat good news is that JD tells me the 6.2um stellar contribution is not dominant on average (though highly variable). sec 5.2.1 Are [SIV]10.51 and [NeV]14.32 too weak to be useful? It might be worth remarking on what fraction of the targets have > 5 sigma detections of these lines. If the fraction is small, then it is clear that they are not very useful as diagnostics. Since Genzel et al used NeV, it might be good to make this point, since it then emphasizes the value of [SiII]34.82. --We have only 2 NeV detections out of 50 nuclei and 26 extranuclear regions. The numbers are 17+7 for [SIV]. There are more [OIV] detections, and that is one reason I chose to use it in Figure 3. You make a good suggestion; I have added text indicating that these lines are less likely to be detected. sec 5.2.4 gives 10^3-10^4 cm^{-3} as "expected from pure HII regions". I'm not sure that this is a fair expectation. The paper cited in support of this (Wang et al 2004) made observations with a 2" slit, and of course targeted high surface brightness HII regions, for which densities 10^3-10^4 cm^{-3} are found. However, in a ~20" arcsec aperture on a galaxy 10 Mpc away, the ~1 kpc aperture will include a lot of more diffuse emission -- lower density HII regions, and even the very diffuse emission from what in the Milky Way we call the Reynolds layer. In the Milky Way the diffuse ionized medium accounts for about 10% of the total recombinations -- not dominant, but perhaps enough to affect the density estimates. --I agree, and that's why I added the qualifier "uncontaminated by the surrounding neutral ISM." However, in light of your comment I have added "on kiloparsec scales" to emphasize the effects of our large apertures. Table 4: last two rows should be for regions V and VI, not IV and IV --Fixed Figs 2,3,4,5,6 -- Should state whether the error bars are 1 sigma or 2 sigma (or whatever). --Fixed Fig. 6 -- three points fall in the forbidden zone (below the low density limit). Is this consistent with statistical fluctuations? Probably if the error bars are 1 sigma, but maybe not if they are 2 sigma. In that case one might ask whether the ratio of collision strengths might be incorrect. It might be a good idea to have a label pointing to the dotted line (Seyfert+Liner Avg) and another to the solid line (HII Avg) -- as the figure is now they don't stand out (alternatively, the line weight might be increased for these two lines so they will be noticed). --These are 1sigma error bars. I have increased the line weights.