---
title: Sieve, Laser, Camera: Why Three Answers?
url: https://oosioo.com/en/p/%EC%B2%B4-%EB%A0%88%EC%9D%B4%EC%A0%80-%EC%B9%B4%EB%A9%94%EB%9D%BC%EB%8A%94-%EC%99%9C-%EB%8B%A4%EB%A5%B8-%EB%8B%B5%EC%9D%84-%ED%95%98%EB%82%98
date: 2026-10-14T13:34:53+00:00
author: SYSOP
summary: Measure coffee grounds with a sieve, a laser and a camera and you get three answers. Running virtual grounds with known sizes through all three, answers that agree within 1% for spheres split into 547, 691 and 977 µm for coffee-shaped particles. What laser diffraction really measures, and how far the belief that 'the laser is the right answer' holds.
---
# Sieve, Laser, Camera: Why Three Answers?

The last two posts settled two questions to ask before reading any grind-size number: what was called the diameter, and what the percentages were percentages of. This time we look at the machines that produce the numbers.

There are three common ways to size coffee grounds. You can sieve them, shine a laser through them, or photograph them. Measure the same grounds all three ways and you get three answers. Which one is right? The usual reply is "the laser." It is a lab instrument, it is expensive, and it returns a smooth curve, so the belief comes easily. In this post I build virtual coffee grounds whose true sizes I know, push them through all three instruments, and test that belief.

## Same grounds, three answers

The particle-sizing company Microtrac has published an application note that measures coffee grounds by several methods. The same grounds went through an image analyzer (a CAMSIZER X2, which films the powder with two cameras as it falls) and a laser diffraction instrument. Image analysis found particles up to 2 mm long, while the laser result contained nothing larger than 1,200 µm. On the other hand, the median (d50) and the share of fines below 200 µm agreed well between the two. The note explains that laser diffraction returns an average over the width and length of each particle: it comes closest to the image analyzer's circle-equivalent diameter, but with a wider spread.

Sieve results matched the image analyzer's "width." The note also states that for irregularly shaped particles such as coffee, a reasonable and reliable comparison between sieve analysis and laser diffraction can hardly be achieved. This is a vendor document, and it does not print the median values. In my search I found no peer-reviewed paper that measured the same coffee by sieve and by laser and reported the difference in numbers.

So even the instrument makers agree that the three methods disagree. This post calculates where the difference comes from and how large it is.

## The laser measures the pattern of a shadow

Sieves and photos were covered in earlier posts. Laser diffraction is new here, so let us start with how it works.

![A set of grey lab instruments. The large central unit has MASTERSIZER written on its side; on the left is a powder feeder with a transparent cover, and on the right a unit stirring liquid in a beaker, connected to the main body by tubes](/uploads/6546c5d5977b0c5a.webp =720)

*A laser diffraction particle size analyzer (Malvern Mastersizer 3000). The optical bench is in the middle; the units on either side disperse the powder in air or water and feed it through. Photo: Plogeo / Wikimedia Commons (CC BY-SA 4.0, cropped)*

The powder is dispersed in air or water and carried through a laser beam. Light grazing the edge of each particle bends and spreads a little. That is diffraction. Behind the particles, a detector receives a pattern of concentric rings.

![A board lit by green light showing concentric rings on a dark background: a bright central disc surrounded by many alternating bright and dark rings](/uploads/a45a31d630ced610.webp =560)

*A diffraction pattern made by a laser passing through a small pinhole. A single particle makes the same kind of pattern (an opaque disc the size of the hole produces the same rings away from the centre). Photo: Anaqreon / Wikimedia Commons (CC0)*

The key to reading size from this pattern is simple. The bigger the particle, the less the light bends. The first dark ring appears at an angle of roughly 1.22 × wavelength ÷ diameter. With the red laser common in these instruments (wavelength 0.63 µm), a 1 mm particle puts its first dark ring at 0.04°. A 30 µm fine puts it at 1.5°.

![The three panels on the left are computed diffraction patterns: concentric rings for a 500 µm sphere, rings twice as widely spaced for a 250 µm sphere, and elliptical rings elongated at right angles to the long axis of an 800 × 400 µm elliptical flake. On the right, a plot of intensity against scattering angle (0.01–20°, log scale): the first dark ring (dotted) falls at 0.04° for 1,000 µm, 0.15° for 300 µm and 1.5° for 30 µm](/uploads/d7eb0154ab36cf52.webp =760)

Halve the particle and the ring spacing doubles. An elongated flake makes an elliptical pattern, narrow along the flake's long direction and wide along its short direction, because the pattern is the Fourier transform of the shadow. Any vision engineer will recognise the relationship.

Real grounds hold millions of particles of different sizes, so their patterns all pile up on the detector. The instrument unravels the sum. It precomputes the patterns of spheres of many diameters and searches for the mix of spheres whose summed pattern matches the measurement. The international standard for laser diffraction, ISO 13320, defines this method. It gives the applicable range as roughly 0.1 µm to 3 mm, and it defines the result as a size distribution of the "volumetric sum of spherical particles" whose predicted pattern matches the measured one. The standard states the reason itself: the technique assumes a spherical particle shape in its optical model.

The instrument's upper size limit comes from the same angles. Malvern's Mastersizer 3000 is specified up to 3,500 µm, and its innermost detector sits at 0.015°. The first dark ring of a 3.5 mm particle is at about 0.013°, right at that edge.

Two things follow. First, the laser's answer is a sphere size from the start: even when particles are not spheres, it answers with "the spheres that make the same pattern." Second, its answer is volume-weighted. As the previous post showed, a sieve counts by mass and a photo counts by number.

## One particle, four rulers

Put one flat coffee flake into each instrument. What size does each one assign? I worked it out for a flake 1,000 µm long, 710 µm wide and 355 µm thick. A width-to-length ratio of 0.71 is the median of the particles measured in Gagné's example photo; thickness = half the width is the assumption used in earlier posts.

![On the left, two brown ellipses: the top view, 1,000 µm long and 710 µm wide, and the side view, 355 µm thick. On the right, a horizontal bar chart of the size assigned to the same particle: sieve 561 µm (smallest square hole it slips through), equal-volume sphere 632 µm (the size that matters for mass), laser 682 µm (sphere with the same average tumbling shadow), camera 843 µm (circle with the area of the resting outline)](/uploads/4308944a44cf6e98.webp =760)

- **The sieve says 561 µm.** That is the smallest square opening the flake can slip through with its smallest cross-section turned to the hole. As we saw in "What Is a Diameter?", a flat flake tilted diagonally passes through holes smaller than its width
- **The camera says 843 µm.** That is the diameter of the circle with the same area as the flake lying flat on paper. Thickness never shows up in a photo
- **The laser says roughly 682 µm.** A flake in the flow tumbles in every direction, so the shadow the light sees keeps growing and shrinking. The average shadow of a convex body over all directions is a quarter of its surface area (Cauchy's theorem). The sphere with that same average shadow is 682 µm across. In the powder experiment below, the laser's answer stayed within 3% of this reference
- **The equal-volume sphere is 632 µm.** This is the size that matters when you think in mass

One flake picked up four numbers from 561 to 843, a factor of 1.5. None of the instruments is wrong. Each holds up a different ruler. If the particle were a sphere, all four numbers would be the same.

## Putting virtual grounds through three instruments

What about a whole powder rather than one particle? I could not get three real instruments, so I built them in the computer.

First, virtual grounds whose true sizes are known. I reused the model with fines from the previous post. The coarse peak extends the real particle distribution from Gagné's photo; the fines peak sits at 35 µm by volume and holds 20% of the mass. Each particle gets an elongation (width ÷ length) drawn at random from the 1,578 values measured in the photo, and a thickness of half its width. I drew 6,000 particles and weighted them so that fines and coarse particles carry mass in their real number proportions.

The three instruments work like this.

- **Sieve**: a stack of 13 standard apertures (45 µm to 2.8 mm). Each particle, in its most favourable orientation, falls to the finest sieve it can pass. The mass left on each sieve is weighed
- **Laser**: each particle is tumbled to a random orientation to get its elliptical shadow, and that shadow's diffraction pattern (Fraunhofer approximation) is added onto 64 ring detectors between 0.01° and 30°. The summed pattern is then inverted into spheres, as a real instrument does, with mild smoothing to keep the solution from oscillating
- **Camera**: each particle lies on its broad face and its circle-equivalent diameter is measured. Particles smaller than 330 µm are invisible, the limit of Gagné's example photo. Volume is computed as coffeegrindsize does, assuming the hidden axis equals the short axis

The same grounds went into all three, and I plotted the cumulative curves by mass. For comparison, I also ran a powder in which every particle is a sphere.

![Two cumulative distribution plots. The x-axis is particle size (5–4000 µm, log scale) and the y-axis the mass fraction below that size. On the left, 'if every particle were a sphere', the equal-volume sphere (black dashed), sieve (green dots and line) and laser (blue) almost coincide, with medians of 829, 823 and 832 µm; only the camera (orange) starts at 330 µm and gives 977 µm. On the right, 'coffee-shaped', the curves spread apart: sieve 547, equal-volume sphere 622, laser 691, camera 977 µm](/uploads/b17bc368927baf30.webp =760)

| Instrument | Spherical powder | Coffee-shaped powder |
|---|---|---|
| Truth (equal-volume sphere) | 829 µm | 622 µm |
| Sieve | 823 µm | 547 µm |
| Laser | 832 µm | 691 µm |
| Camera (≥ 330 µm) | 977 µm | 977 µm |

With spherical particles, sieve, laser and truth agree within 1%. Only the camera reads high, because it cannot see anything below 330 µm and takes the median of what is left. That was the problem in the previous post. If the camera could see every particle, it would read 829 µm too.

Once the particles took on a coffee shape, the answers split: sieve 547 µm, laser 691 µm, camera 977 µm. The largest is 1.79 times the smallest. The instruments did not change; only the particle shape did. The main reason the three instruments disagree is not instrument error. It is that coffee is not made of spheres.

## Is the laser the right answer?

So what about the laser? On these virtual grounds it answered 691 µm, 11% above the mass-based truth of 622 µm. What the laser reports is not the sphere of equal mass but something close to the sphere of equal average shadow. The median of that size is 671 µm, 3% from the laser's answer.

The bigger problem is thickness. I repeated the experiment while changing the thickness that photos cannot see.

![A line plot with assumed thickness / width (0.2–1.0) on the x-axis and volume median (300–1200 µm) on the y-axis. The camera (≥ 330 µm, solid orange) stays flat at 977 µm, and the camera that sees every particle (dotted orange) at 829 µm. The laser (blue) rises a little, from 680 to 794 µm; the equal-volume sphere (black dashed) rises steeply from 458 to 783 µm; the sieve (green) rises from 502 to 694 µm](/uploads/d556b90a23291c17.webp =720)

As thickness went from 0.2 to 1 times the width, the mass-based truth moved from 458 µm to 783 µm, 71%. The sieve moved from 502 µm to 694 µm, 38%. The laser moved only 17%, from 680 µm to 794 µm. Below half the width, the laser value barely changes at all. The camera does not see thickness, so it does not move.

The laser is this insensitive to thickness because of shadow area. A thin flake tumbling through the beam shows its broad face, or a slanted face, to the light most of the time; the moments when only its edge faces the beam are brief. So the average shadow hardly shrinks as the flake gets thinner, while mass is directly proportional to thickness. The laser returns a smooth, repeatable curve, but it cannot tell you the mass distribution of flat particles.

The laser curve also grows a longer upper tail. The mass-based D90 is 1,256 µm, but the laser reports 1,521 µm. When an elongated flake passes with its broad face to the light, it looks like a large sphere, and that appearance adds to the coarse end of the distribution. This matches the direction of the Microtrac note's remark that the laser distribution has a wider spread.

That laser results for non-spherical particles differ from other methods is written into the standard. ISO 13320 states that for non-spherical particles the distribution differs from those obtained by methods based on other physical principles, such as sedimentation and sieving. Outside coffee there are numbers too. Konert and Vandenberghe (1997), measuring soils rich in platy minerals, reported that the 2 µm boundary defined by sedimentation (the pipette method) corresponds to 8 µm by laser. In a Malvern application note, three volcanic ashes collected between 90 µm and 125 µm sieves gave laser D10 values of 65–88 µm and D90 values of 175–225 µm. Particles gathered between 90 and 125 µm sieves appeared to the laser spread over 65–225 µm, a spread Malvern attributes to particle shape. That is the same direction as the long laser tail in our virtual experiment.

Repeatability and accuracy are different things. A smooth laser curve that comes back the same on a second run means the laser applies its own definition consistently. It does not mean that definition is the size we want to know.

## Who loses the fines?

Besides the median, I compared the share of fines: how much of the mass passes a 90 µm sieve.

![Horizontal bar chart of the mass share of fines below 90 µm: truth (virtual sample) 19.7%, laser 19.4%, sieve (well dispersed) 19.9%, sieve (if 30% of fines cling) 13.9%, camera (≥ 330 µm only) 0](/uploads/c13f6581bdb45d54.webp =720)

If the powder is well dispersed, both sieve and laser measure the fines share almost exactly. Fines are small, so even with odd shapes little mass crosses the 90 µm boundary. The camera cannot see anything below 330 µm, so its fines share is zero.

The sieve has another weakness. Freshly ground coffee carries static charge. The US Pharmacopeia's sieving chapter (USP <786>) allows adding 0.5% of an antistatic agent when electrostatic charge is a problem, and tells the analyst to confirm that what stays on a sieve is single particles rather than aggregates. The Microtrac note likewise says coffee oil causes clumping, adhesion to the sieve frame and clogging, and recommends air-jet sieving below 200 µm. Charging during grinding has been measured: Méndez Harper et al. (2024) found that grinding dry beans can produce up to 120 nC of negative charge per gram, and that adding just 5 µL of water per gram of beans immediately reduced the electrostatic clumps of fines and coarse particles. Fines that cling to coarse particles or to the mesh are weighed by the sieve as coarse mass. I found no measurement of how much this amounts to. If 30% of the fines cling, the sieve's fines share drops from 19.9% to 13.9%. The 30% is only a scenario; the real fraction will depend on the beans, the humidity and how the sieve is shaken.

The laser has the same problem if the powder is not dispersed: a clump of fines looks like one large particle. That is why laser instruments disperse powder with compressed air or ultrasound. Push too hard, though, and the particles break. Either way, the sample preparation becomes part of the result.

## Is there a right method?

No. More precisely, each method gives the right answer to its own question.

- The sieve answers "how much of this powder passes a hole of so many µm." That is close to the question you ask about filter or basket holes and the gaps in a puck
- The laser answers "how much light does this powder block and scatter." Shadow area is tied to surface area, which makes it useful when thinking about the area water touches
- The camera answers "how broad is each particle lying down," and it keeps the shape of every particle

If you want the "true size" counted by mass, all three instruments share a blank: thickness. Sieve and laser feel it indirectly but do not report it; the camera does not see it at all.

So here are rules you can actually use.

1. **Do not compare numbers from different instruments directly.** A laser D50 of 700 µm and a photo D50 of 700 µm may describe different grounds
2. **Changes measured on the same instrument are trustworthy.** When you move the grinder one step, the curve should move the same way whatever the instrument, though how far it moves will differ between instruments
3. **To convert, measure shape.** Know thickness ÷ width and the calculation in this post lets you move between instruments. Without it, any conversion rests on an assumption
4. **Write the instrument, the diameter definition and the weighting next to every number.** For example, "D50 = 691 µm (laser diffraction, volume-weighted)"

## Try it yourself

In the simulation below you can run the same virtual grounds through all three instruments. On the left are the cumulative mass curves; on the right, each instrument's volume median and fines share below 90 µm. Every time you move a slider, the browser recomputes the laser's diffraction pattern and inverts it. To keep it fast it uses only 3,000 particles, so its numbers may differ from those in the text by a few µm.

Press "all spheres" and the curves collapse onto each other; go back to "coffee flakes" and they spread apart. Lower thickness / width to 0.2 and watch how differently the black dashed line (truth) and the blue line (laser) move. Raise the fines left clinging on the sieve and the bottom of the green curve sinks.

```map /embed/coffee-instruments-2026-en.html 16:12
Simulation measuring the same virtual coffee grounds with a sieve, laser diffraction and a camera. On the left, cumulative curves with particle size (5–4000 µm, log scale) on the x-axis and the mass fraction below each size on the y-axis: truth (equal-volume sphere, black dashed), sieve (green dots and line), laser diffraction (blue) and camera (orange), with the small sizes the camera cannot see shaded. On the right, bar charts comparing each instrument's volume median and the mass share of fines below 90 µm. You can change particle shape (coffee flakes or all spheres), thickness / width, fines mass share, the smallest size the camera sees, fines left clinging on the sieve, and sieve shaking (long enough or short).
```

## What this model leaves out

- **No real instrument was used.** Sieve, laser and camera are all models in a computer. Only measuring the same grounds on all three instruments will show how well this calculation holds
- **The laser uses the Fraunhofer approximation only.** Real instruments use Mie theory, which includes the particles' refractive index and absorption. The difference is small for coarse particles, but for fines below a few tens of µm the result can depend on the refractive index setting
- **Particles are smooth ellipsoids.** Real coffee flakes have edges and rough surfaces, which shift both shadow area and sieve size a little
- **Particles in the laser beam tumble at random.** Elongated particles in a flow tend to line up, so the real average shadow may differ from this calculation
- **Thickness is still an assumption.** Most numbers in this post assume thickness = half the width. The thickness plot above shows how far they move with it
- **The 30% of fines clinging to the sieve is an assumed number,** not a measured one

## Summary

- Laser diffraction inverts the pattern scattered by the particles into the patterns of spheres, reporting a volume-weighted size of "spheres that make the same pattern." Bigger particles bend light less (first dark ring ≈ 1.22 × wavelength ÷ diameter)
- For one flake 1,000 µm long, 710 µm wide and 355 µm thick, the sieve assigns 561 µm, the equal-volume sphere is 632 µm, the laser about 682 µm, and the camera 843 µm
- Running virtual grounds with known sizes through three instruments: with spherical particles, sieve, laser and truth agreed within 1%. With coffee-shaped particles, they split into sieve 547, laser 691 and camera 977 µm, a factor of 1.79
- The laser answered 11% above the mass-based truth (622 µm). As thickness went from 0.2 to 1 times the width, the truth moved 71% but the laser only 17%
- When the powder is well dispersed, sieve and laser both measure the fines share almost exactly. Fines clinging on the sieve, or too little camera resolution, lose them
- There is no right method. Each method answers its own question. Do not compare numbers from different instruments directly, and write the instrument, the diameter definition and the weighting next to every number

"The laser is the right answer" is half true. The laser is an instrument that applies its own ruler consistently. What that ruler measures is a separate question.

## References

- ISO 13320:2020. *Particle size analysis — Laser diffraction methods* — range about 0.1 µm–3 mm, sphere assumption and volume-based result, non-spherical particles differ from sieving and sedimentation (clauses 1, 4.2). [Official sample](https://cdn.standards.iteh.ai/samples/69111/ae231f15f03341da9699f3ca7d3e8dcd/ISO-13320-2020.pdf)
- Microtrac. *Coffee powder* (application note, instrument manufacturer). [Link](https://www.microtrac.com/applications/coffee-powder/) — image analysis finds particles up to 2 mm long, laser shows none above 1,200 µm, d50 and share below 200 µm agree, sieve matches width, sieve and laser hardly comparable for irregular particles
- Malvern Panalytical. *Mastersizer 3000* brochure MRK1872-07 — range 10 nm–3,500 µm, He-Ne 632.8 nm, detector angles 0.015–144°
- Malvern Panalytical. *Beyond the sieve* (application note AN250722, manufacturer) — laser D10 65–88 µm, D50 124–131 µm, D90 175–225 µm for three volcanic ashes from a 90–125 µm sieve fraction
- Konert M, Vandenberghe J (1997). Comparison of laser grain size analysis with pipette and sieve analysis: a solution for the underestimation of the clay fraction. *Sedimentology* 44, 523–535. [doi:10.1046/j.1365-3091.1997.d01-38.x](https://doi.org/10.1046/j.1365-3091.1997.d01-38.x) — pipette 2 µm ≈ laser 8 µm (abstract checked)
- Méndez Harper J, et al. (2024). Moisture-controlled triboelectrification during coffee grinding. *Matter* 7, 266–283. [doi:10.1016/j.matt.2023.11.005](https://doi.org/10.1016/j.matt.2023.11.005) — up to 120 nC/g; 5 µL/g of water reduces electrostatic clumps
- United States Pharmacopeia. <786> Particle size distribution estimation by analytical sieving — 0.5% antistatic agent; confirm that sieve residues are not aggregates
- Slepian Z (2011). The average projected area theorem. [arXiv:1109.0595](https://arxiv.org/abs/1109.0595) — average shadow of a convex body = ¼ of its surface area (Cauchy)
- Gagné J. *coffeegrindsize*. GitHub. [Link](https://github.com/jgagneastro/coffeegrindsize) — example photo; hidden-axis assumption for volume

## Image credits

- Laser diffraction particle size analyzer: Plogeo, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:The_Mastersizer_3000.jpg), [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) (cropped)
- Laser diffraction pattern: Anaqreon, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Airy_disk_created_by_laser_beam_through_pinhole.jpg), CC0 (cropped)
- All other figures and the simulation: made by the author. Particle elongations were measured from the example photo in Jonathan Gagné's [coffeegrindsize](https://github.com/jgagneastro/coffeegrindsize) (MIT licence)