This paper reports on a laboratory study of graupel-graupel and graupel-snow aggregate collisions. The numbers of fragments and the size distributions produced are presented. 

The authors do a good job of describing their setup and show some compelling results. It is very good to see these new observations coming out that can eventually be implemented in numerical cloud models, and the authors have provided representations of the fragment particle size distributions as well and fragment number-collision energy relationships.

This is hopefully the beginning of a suite of experiments to explore the behaviour of these collisions for a range of graupel and aggregate sizes under different temperature and saturation conditions, that will be reported in future papers.

I think the paper is publishable. I have some very minor comments.

Main:

I would suggest that highly dendritic growth on a graupel particle seen in fig 5b would not be observed in the real atmosphere. It would be difficult to obtain saturations above water saturation (~14% at -14C). At water saturation there would be droplets that would continue the riming.

For a realistic graupel with vapour grown surface ice i think that a superaturation with respect to ice but below water saturation is required.

It would be good to caveat the results for the dendrite-covered_graupel vs graupel collisions.

Perhaps it would be possible to include the bare graupel-graupel collision results that were alluded to?

Minor:

line 215: What was the 'glue' used for sticking? Were the crystals just brought together at ice saturation or slightly above? How long were the crystals allowed to sinter for? I imagine there will be sensitivity to this. In the results the production process for the aggregates is mentioned, but perhaps it is worth saying that this is something that could be explored more systematically in the future?

line 302: Fig 12 i think  needs a caveat to mention that these results are likely to be an upper bound because of the very high saturations the graupel were exposed to. Much more than is likely in a real cloud.

line 344: For fig. 14 graupel-snow collisions it may be appropriate to just suggest a  mean and range (e.g. 200 splinters ranging from 100-400 to capture 95% of measurements. Hopefully, later experiments will provide enough data to parameterise the degree of separation effect. For models some average would likely be necessary to use.

line 350: The size distribution of fragments is very welcome. For implementation in a model this would likely need to be scaled by the size of the snowparticle being collided with.

In the first instance the use of 10mm snow aggregates could be used to scale the x axis (at least as an extra axis)?

line 354: Only 16 distributions in here so difficult to draw too many conclusions from the individual modes - apart from the mode at 50um being the size of the monomers.

line 368: This 50um mode is just the size of the monomers used to construct the aggregate, so unless Vardiman constructed their snow in the same way there is unlikely to have been a mode in those observations?

line 401: Agreed. It would be great to see results for a range of graupel sizes and snow sizes to cover the phase space required in a numerical cloud model.