---
title: Do Frozen Beans Grind Finer?
url: https://oosioo.com/en/p/%EC%96%BC%EB%A6%B0-%EC%9B%90%EB%91%90%EA%B0%80-%EB%8D%94-%EA%B3%B1%EA%B2%8C-%EA%B0%88%EB%A6%B0%EB%8B%A4
date: 2026-11-03T06:12:47+00:00
author: SYSOP
summary: I reread Uman et al. (2016), the starting point of the idea that frozen beans grind finer and more evenly. It used one coffee, one grinder, one day and two grinds per temperature, with no statistical test of the temperature effect; on one metric, freezer beans (−19 °C) came out larger than room-temperature beans. The co-authors included people from the grinder maker, the instrument company and the café. Giving a virtual grinder candidate "cold bean" properties, none produced a finer and narrower distribution at once. With a tool for working out how many grinds you need.
---
# Do Frozen Beans Grind Finer?

There is a story that beans frozen and ground straight from the freezer come out finer and more even. A barista champion used beans chilled with dry ice, and more and more people keep single doses in small tubes in the freezer. The story usually starts from one paper published in *Scientific Reports* in 2016: Uman and colleagues (2016).

This time I reread that paper. The short version: the observation that grinding cold gave a narrower, smaller distribution is interesting. But it came from one coffee, one grinder, one day and two grinds per temperature, with no statistical test of the temperature effect. On one of its metrics, freezer beans came out coarser than room-temperature beans. When I gave a virtual grinder candidate properties of "cold beans" one at a time, none of them produced a finer and narrower distribution at once.

## What the paper did

![Looking down into a cryopreservation dewar. A silvery hose is pouring liquid nitrogen, and white vapour fills the dewar. Handwritten labels are stuck around the rim](/uploads/976683289f146b81.webp =520)
*Filling a cryopreservation dewar with liquid nitrogen* · Photo: U.S. Fish and Wildlife Service Southeast Region, Wikimedia Commons, public domain

Uman and colleagues held 20 g of beans at four temperatures for two hours: liquid nitrogen (−196 °C), dry ice (−79 °C), a freezer (−19 °C) and room temperature (20 °C). It is often described as "comparing fridge and freezer", but there was no fridge (4 °C) condition. The beans were ground in a café in Bath, England, on an EK43 (Turkish burrs, 1,480 rpm, fixed dial) and measured with a laser diffraction analyser (Beckman Coulter LS13 320). Results were shown mostly as number-based distributions.

The paper reports that colder beans gave narrower distributions, and that the mode of the number distribution fell by 31% from 20 °C to −196 °C. The biggest change was between 20 °C and −19 °C. Liquid nitrogen and dry ice differed little, and beans frozen in liquid nitrogen and returned to room temperature were no different from room-temperature beans.

![Two panels. On the left, the 'fines cutoff' (a boundary size set from the cumulative number distribution, µm) at four temperatures with ±3 error bars: −196 °C (liquid nitrogen) 61, −79 °C (dry ice) 63, −19 °C (freezer) 73 and 20 °C (room) 70 µm. The freezer is larger than room temperature, and the error bars overlap. On the right, a table of the experiment's set-up: one coffee (the Methods and the figure caption name different origins), one EK43 with a fixed dial, 2 grinds × 3 measurements per temperature, one day, no statistical test for temperature, taste and extraction not measured, number-based distributions (weighted to fines)](/uploads/4cf86f0f67da8397.webp =760)

Another metric in the same figure is less simple. What the paper calls the fines cutoff was 61±3 µm at −196 °C, 63±3 µm at −79 °C, 73±3 µm at −19 °C and 70±3 µm at 20 °C. Freezer beans came out larger than room-temperature beans, with overlapping error bars. Home freezing is the −19 °C case. It is hard to say "freezing makes beans grind finer" from this table alone.

## How many grinds?

The paper says it used "six datasets" per temperature. In practice that is two grinds, each measured three times. Putting the same powder through the analyser three times repeats the measurement, not the grinding. Counted by grinds, it is two per temperature, all from one coffee, one grinder and one day. There was no statistical test of the temperature effect.

The supplementary ANOVA comparing origins had a structural problem. The "94 samples" for each origin were the shares (%) in 94 size bins. Whatever the distribution, the bin shares add up to 100%, so the mean of 94 bins is always 100 ÷ 94 = 1.0638%. The four origins must have the same mean, and p = 1 follows. It cannot support the conclusion that origins were "similar regardless". To see whether distributions differ, you need repeated grinds compared on values such as D50, or a test that compares whole distributions.

How many grinds would have been needed? I used the virtual grinder to get a feel. Running the 3D flat burr from earlier episodes (gap 1.6 mm, 1,000 rpm) eight times with different seeds and 12 beans each, D50 varied by ±3.5% (standard deviation) around 660 µm, with the same grinder, beans and setting. A real 20 g dose holds about 150 beans and would scatter less, but the order in which beans enter and the state inside the grinder differ each time, so the scatter does not vanish.

![Grinds per condition (2–30) across and chance of detecting the difference (%) up. Two orange lines (scatter ±3.5%, virtual grinder with 12 beans): solid for a true difference of 5%, dashed for 2%. Two blue lines (scatter ±1%, if a 20 g dose of ~150 beans), the same way. With ±1% scatter and a 5% difference (solid blue), two grinds already give almost 100%. With ±1% scatter and a 2% difference (dashed blue), two grinds give 52% and four give 81%. With ±3.5% scatter and a 5% difference (solid orange), two grinds give 30% and nine give 86%. With a 2% difference and ±3.5% scatter (dashed orange), even 30 grinds give only 60%. A dotted line marks 80%, and a red vertical line at 2 marks 'Uman et al.: 2 per temperature'](/uploads/6230bf3331dde5db.webp =760)

To catch a 5% change in D50, two grinds are enough if each grind scatters by ±1%. If the scatter is ±3.5%, it takes about nine grinds to catch it eight or nine times out of ten. For a 2% effect at ±3.5% scatter, even thirty grinds catch it six times out of ten. These curves use a normal approximation, so with few grinds they are more optimistic than a real t-test. The 31% change in mode reported by Uman and colleagues is a large effect and may have shown up with two grinds. But to say something about small differences at freezer temperatures, such as between −19 °C and 20 °C, more replicates were needed.

## Who wrote it

The eleven co-authors came from many places: an employee of Mahlkönig, which makes the EK43 used in the experiment; two employees of Meritics, the analysis company that donated the instrument; the two owners of the café where the tests ran; the roaster who supplied beans; the barista who founded Barista Hustle; and university researchers. The paper declares "no competing financial interests" and lists no funding. The corresponding author, Hendon, said in an interview that he worked with Kyle Ramage and Mahlkönig before the 2017 US Barista Championship. Ramage won that championship using beans chilled to −79 °C with dry ice.

Having interests does not make a result wrong. Because people from the trade wrote it together, the experiment could use a real grinder and real beans. But sentences in the conclusion such as "grinding cold should raise extraction efficiency" or "busy cafés tighten the grind in the afternoon because the burrs warm up" were not measured in this paper. Taste and extraction were not tested.

Peer-reviewed studies that have since remeasured the question are rare. Pignatone and colleagues (2026) at the University of Florence warmed or cooled beans just before grinding and measured particle size and espresso compounds; according to the abstract, cooling had weaker effects than warming and reduced caffeine extraction and some aroma compounds. I could not check the exact temperatures or size data in the full text. Skoltech's "15–20% more uniform at −100 °C" in Russia is confirmed only through a press release.

## Why would cold make a difference?

![A close-up of roasted coffee beans. Several dark brown beans show oily surfaces and the groove down the middle](/uploads/5116f196f1b31911.webp =560)
*Roasted coffee beans, close up* · Photo: Jameswasswa, Wikimedia Commons, CC BY-SA 4.0

The usual explanation is that "the water in the beans freezes and makes them brittle". But roasted beans contain almost no water that can freeze. In measurements by Rocculi and colleagues (2011), roasted beans only developed freezable water above a water activity of 0.86, and below that, water actually made them harder. Freshly roasted beans are far drier than that. Uman and colleagues also wrote that it is uncertain what transition, if any, the beans go through. I found no data on how the strength of roasted beans changes with temperature.

So I gave the virtual grinder, one at a time, properties that "cold beans" might plausibly have: weaker breaking force (0.8 and 0.6 times), more uniform strength from bean to bean (Weibull modulus 3 → 12), and breaking into more pieces (2–4 → 3–6 fragments).

![Two charts across five conditions (baseline, strength ×0.8, strength ×0.6, less strength scatter, 3–6 fragments per break). On the left, D50: bars at 660, 627, 509, 658 and 694 µm. On the right, distribution width (D90 ÷ D50): 1.52, 1.56, 1.86, 1.51 and 1.43. Grey dots are individual runs with different seeds; the dotted line is the baseline](/uploads/daf893b19d31a1ae.webp =760)

Weaker beans gave finer grounds. At 0.6 times the breaking force, D50 fell 23%, from 660 µm to 509 µm. But the width (D90 ÷ D50) grew from 1.52 to 1.86: the coarse end (D90) is set by the gap and barely moved, while only the fine end grew. More uniform strength changed almost nothing. The distribution narrowed only when beans broke into more pieces (1.43), and then D50 actually grew coarser, to 694 µm, because beans broke fewer times and shed fewer crumbs.

In this model, no single property gave a "finer and narrower" distribution at once. If that result is real, two things must happen together when beans are cold: they must break more easily and break into smaller pieces. Or something this model lacks must be at work, such as the gap changing as the burrs cool. Feeding cold beans into a grinder cools the burrs a little too.

## Check it yourself

The first tab of the tool below repeats a virtual experiment 200 times. Change the true difference, the scatter per grind and the number of grinds per condition, and you see the distribution of differences the experiments found and the share in which a t-test caught the difference. The second tab shows why an ANOVA on the mean of bin shares always gives the same mean.

```map /embed/coffee-frozen-2026-en.html 16:9
A tool for thinking through experiment design and statistics. In the first tab, 'How many grinds to see a difference', sliders set the true difference (D50 0–20%), the scatter per grind (±1–15%) and the grinds per condition (2–20). The chart on the left shows the distribution of D50 differences seen in 200 simulated experiments: green are experiments where a t-test (5% level) caught the difference, grey are those that missed, and an orange vertical line marks the true difference. On the right, the share of experiments that caught the difference is shown in large type, with a 'Re-draw' button. In the second tab, 'The trap of averaging bins', two differently shaped distributions (blue and orange) are drawn over 94 size bins, showing that their bin means are always the same, 1.0638%. A 'Change distributions' button draws new ones.
```

## What this conclusion leaves out

- **I did not measure the effect of freezing directly.** The simulation here tests "if cold beans had this property". There is no measured data on how the strength of roasted beans changes with temperature
- **The virtual grinder's scatter cannot be carried straight to real grinding.** It comes from 12 beans and spherical particles
- **I could not check the full text of Pignatone and colleagues (2026).** Knowing their temperatures, replicates and size data could change the judgement here
- **Taste is not covered.** Whether coffee ground cold tastes better is a different question, and I found no controlled experiment that answers it
- **Storage is not covered.** The claim that freezing slows the loss of aroma is separate from grind size. Advice to portion and seal beans so they don't collect condensation when taken out is common, but the measured evidence is thin

## Summary

- Uman and colleagues (2016), the starting point of "frozen beans grind finer", used one coffee, one grinder, one day and two grinds per temperature. There was no statistical test of the temperature effect, and no fridge (4 °C) condition
- The mode fell 31% from 20 °C to −196 °C, but the fines cutoff of freezer beans (−19 °C) at 73 µm was larger than at room temperature (70 µm), with ±3 µm error bars
- The supplementary origin ANOVA compared means of bin shares that always sum to 100%, so p = 1 by construction
- The co-authors included people from the grinder maker, the instrument company, the café and the roaster. The sentences about taste and extraction efficiency are speculation, not measurement
- In a virtual grinder, weaker beans gave finer but wider distributions, and beans that shattered into more pieces gave narrower but coarser ones. No single property gave "finer and narrower" at once
- If D50 scatters by ±3.5% at the same setting, catching a 5% difference takes about nine grinds

## References

- Uman E, Colonna-Dashwood M, Colonna-Dashwood L, Perger M, Klatt C, Leighton S, Miller B, Butler KT, Melot BC, Speirs RW, Hendon CH (2016). The effect of bean origin and temperature on grinding roasted coffee. *Scientific Reports* 6, 24483. [doi:10.1038/srep24483](https://doi.org/10.1038/srep24483) — main text and supplement
- Pignatone M et al. (2026). Effects of coffee bean thermal treatments on particle size distribution and espresso bioactive compounds. *Applied Sciences* 16(10), 4886. [doi:10.3390/app16104886](https://doi.org/10.3390/app16104886) — abstract only
- Rocculi P et al. (2011). Water activity and texture of roasted coffee beans (hardness, elastic modulus, freezable water). *Journal of Agricultural and Food Chemistry*. [doi:10.1021/jf201333a](https://doi.org/10.1021/jf201333a) — abstract only; title not checked against the original
- Little Black Coffee Cup, interview with Christopher Hendon — mentions the 2017 US Barista Championship collaboration
- Cundall PA, Strack ODL (1979). A discrete numerical model for granular assemblies. *Géotechnique* 29(1), 47–65. [doi:10.1680/geot.1979.29.1.47](https://doi.org/10.1680/geot.1979.29.1.47)

## Image credits

- Filling a cryopreservation dewar: U.S. Fish and Wildlife Service Southeast Region, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Filling_Cryopreservation_Dewar_with_Liquid_Nitrogen_%285544592746%29.jpg), public domain
- Roasted coffee beans: Jameswasswa, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Macro_close_up_look_of_roasted_coffee_beans.jpg), CC BY-SA 4.0
- All other charts, the cover and the tool: made by me

