The article reads really well I think. I guess the main weakness of the paper 
is the small size of the sample which strongly constrains what can be 
concluded in the end. I just have small comments.

* In the introduction, maybe it would be adequate to talk a bit about cold flows.

I'm familiar with cold flows in the galaxy cluster context, but not within galaxies.  Feel free to provide some text here.

* Still in the introduction, is there anything on M31 from PHAT?

There was a semi-relevant conference proceeding from Gouliermis in 2014.  The title of a paper submitted last year by Lewis et al. sounds promising, but it hasn't appeared on astro-ph yet.

* p3 "the bulk [...] are" --> is

Fixed

* I feel the referee may be a bit annoying regarding the sample. In the end, 
what is the selection function? What are the biases of this sample? Why have 
these galaxies been selected rather than others?

I expanded a little on the sample selection, namely targets easily observable during Summer 2014 and preferentially with good UV and IR ancillary data.

* In Sect. 3.1, I guess the 2 kJy/sr sensitivity is per pixel?

Yes.  I have added this fact to the text.

* I won't change anything anyway, but I would not use a gaussian kernel to 
degrade to the MIPS 24 PSF. I would rather use one of Gonzalo's kernels.

* Sect. 4, "relatively course" --> coarse

Thanks

* Sect. 4.2, CIGALE has changed a lot since the Noll days. I suggest to also 
mention Boquien et al. and Burgarella et al. (both in prep.).

Thanks

* Regarding the input parameters in Table 3, beta is not the UV slope. It is the 
slope modifier of the attenuation law (beta=0 --> starburst law, beta=-0.5 ~LMC law). 
I think it would be useful to indicate the number of values for each parameter 
to give a sense of the density of the sampling.

Thanks -- I have fixed that description, and explicitly added the grid search values.

* Later in Sect. 4.2, the PDF is basically a probability-weighted mean 
(P propto exp(-chiČ/2)

I added "probability-weighted" before "mean".

* Sect. 5.2, if I am not mistaken, all galaxies that have a displaced IR peak 
are pretty late type. I would have expected that for early-type galaxies 
rather as their inner regions are quite depleted of gas and dust (think M81 or 
M104 for instance).

It must be that the displaced peaks are sites of more current star formation.  This is in fact seen in the radial tau plots.  I've added this to the discussion.

* Maybe it would be useful to accompany Fig. 4 of a statistical test to show 
that the trend is significant.

Maybe in the next round of revisions.

* I was just thinking, there must be theoretical profiles available in the 
literature. Maybe you could try to compare them quantitatively with the 
observed ones and see whether anything can be drawn from that.

This could be an excellent opportunity for you.  ;)