What pressure actually reaches the coffee

Nine bar is the most repeated number in espresso and one of the least examined. Ask where it comes from and you get a confident story about Achille Gaggia's spring-lever machines; ask for the document behind the story and there isn't one. Ask what the number refers to and it turns out to name at least three different pressures that are not equal during a shot. And as of 2026 there is a fourth problem, which is the most interesting of them: the coffee bed is not a fixed resistance that some pressure is applied to. Its permeability depends on the pressure applied to it, so above roughly five bar, raising pressure stops buying proportional flow.

This article separates what is documented from what is merely repeated, and it will end up concluding less than most pages on the subject do.

Three pressures, one number

On a typical prosumer machine there are three distinct quantities.

Pump output pressure. What the pump generates, bounded by the over-pressure valve. On rotary-pump machines the panel gauge is often a pump gauge, and it reads its set pressure whenever the pump runs — including with no portafilter in the machine at all. That behaviour is a useful diagnostic: a gauge reading nine bar with nothing installed is telling you about the pump, not about coffee.

Gauge pressure. Where the sensor actually taps in. On E61 machines the brew gauge is commonly plumbed into the channel feeding the dispersion screen, frequently at the end of a narrow tube, and typically before the gicleur, the flow-restricting jet in the group. Anything downstream of the tap point is invisible to it.

Puck inlet pressure. What the water is at when it meets the coffee. This is the only one of the three acting on the bed, and on most machines nothing measures it.

Between the tap point and the basket sit real restrictions: the gicleur, and on a classic E61 the preinfusion chamber as well. Each imposes a pressure drop.

How big is the drop? Nobody who builds these machines says

This is the second undocumented number in the article, and it is undocumented in a more surprising way than the first.

The restriction itself is precisely specified. Parts suppliers list the E61 jet by thread and orifice — 0.6 mm and 0.7 mm versions are both sold as standard replacement parts. What no manufacturer, distributor or service document appears to publish is what that orifice does to the pressure.

La Marzocco's own 62-page GS/3 installation and operation manual is the clearest illustration. Searched in full, it contains no instance of gicleur, gigleur, restrictor, orifice or pressure drop. It specifies the endpoints and skips what is between them: "The water pump is factory set at 9 bar pressure", and "When brewing, the coffee boiler pressure should be between 8-10 bar." The reader is told what to set and what the gauge should read, and nothing about why those are different from what the coffee gets.

So the figure in circulation comes from users. The most-quoted version is a single post by cafeIKE on home-barista.com, 11 September 2021: "Typically e61 brew pressure gauges are pre gicleur. 10 on the gauge can be 8.5 on the puck." That is where the familiar 1.5 bar comes from — one sentence, about one machine, offered as a rule of thumb rather than a measurement.

Other practitioners report smaller gaps on other machines, and the reports taken together support something like one to two bar, depending on the machine. They do not support 1.5 bar as a figure, and this article previously used it as though they did. Anyone quoting a single number here is quoting one person's machine.

The dealer practice everyone cites is better attested, because it is a report of what someone did rather than a measurement. From the same thread, the owner who started it: "Both of my E61 machines with Flow Control devices have their overpressure (aka bypass) valves set to 10 bar by their respective dealers. Both dealers say they do this to achieve a brew pressure of 9 bar." That establishes the practice and the dealers' stated reasoning. It does not independently establish that the offset is correct.

Why a blind basket hides the problem

The behaviour that makes this concrete: with a blind basket inserted, the pump gauge and the brew gauge read approximately the same. During an actual extraction they separate.

The mechanism is that pressure drop across a restriction is a function of flow. With a blind basket there is no flow, so there is no drop, and every point in the system settles to the same pressure. Start a real shot and water moves; each restriction now consumes some pressure, and the puck sees less than the gauge reports.

This inverts a common habit. Setting brew pressure against a blind basket calibrates the system in the one condition where the number you are trying to control cannot be wrong. It is a reasonable way to set the over-pressure valve. It is not a measurement of what the coffee experiences.

It also means puck inlet pressure is not constant during a shot. As the bed's resistance changes — swelling, fines migration, channel formation — flow changes, so the pressure drop across every upstream restriction changes with it.

A note on why the drop happens

Free explanations of this commonly attribute the pressure drop across a restriction to Bernoulli's principle. That is the wrong mechanism. Bernoulli describes inviscid flow, where pressure and velocity trade off with no losses; it does not produce a permanent drop. What produces one here is viscous loss — Darcy–Weisbach through a jet, and Darcy's law through a packed bed. The distinction is not pedantry: a Bernoulli account predicts pressure recovers once the flow slows again downstream, and it does not.

The puck is not a fixed resistance

The three-pressures framing still contains an assumption worth dropping: that the bed is a fixed resistor and the only question is what pressure arrives at it.

Waszkiewicz and colleagues tested that directly on a café-grade machine and published the result in Physics of Fluids in June 2026, with a free preprint and their data and analysis code public. Below about five bar the puck behaves as Darcy's law says it should — flow rises roughly linearly with pressure. Above that it does not. Flow saturates: pushing harder yields progressively less additional flow, and at long times the rate can even decline slightly toward a stable value.

The mechanism is poroelastic. The bed is not rigid. Pressure compacts it, compaction reduces porosity, and reduced porosity raises resistance — so part of any increase in pressure is spent making the bed harder to push through. Their minimal model captures this with two calibration constants, a calibration pressure of 12 ± 3 bar and a calibration flow of 1.90 ± 0.15 g/s, and an estimated equilibrium porosity of about 0.122.

Two things follow for anyone reasoning about pressure at the puck.

The first is that the puck is not a fixed load. Its resistance is a function of the pressure applied to it, which makes the bed part of the hydraulic circuit rather than something attached to the end of it. Pressure at the puck inlet remains a real and measurable quantity — but it is not a setting, and on its own it does not determine flow, because the same pressure meets a denser bed the longer it is held.

The second is that the returns on high pressure are not what a linear intuition suggests. Going from five bar to nine is an eighty per cent increase in pressure, and it does not buy eighty per cent more flow: five bar is already at the edge of the saturating regime. This is a mechanism-level reason to hold the number loosely, and it is independent of, and considerably stronger than, the historical argument below.

Where the number came from, as far as anyone can show

Two claims travel together here and they have very different evidential standing. Pulling them apart is the main thing this article has to offer.

"Nine bar became the convention" is documented. It is in machine specifications, in trade standards, and in the peer-reviewed literature as the stated norm. Nobody disputes it.

"Gaggia's lever spring produced nine bar, and that is why" is not documented anywhere this article could reach. It is an engineering claim about a specific mechanism, and it is asserted without a patent, a drawing, a specification or a measurement behind it.

The trail is worth walking, because where it stops is informative. Gaggia's own corporate history gives a patent number for the 1938 "Lampo" system — 365726, filed 5 September 1938 — which shows the company does cite its patents when it has them. For the 1947 spring-lever machine, the machine the whole story rests on, the same history gives no patent number and no pressure figure at all, saying only that it forced hot water at high pressure through the coffee. The one Gaggia patent that is freely retrievable in full is US2898844A, "Machines for infusing coffee and like liquid extracts", inventor Giovanni Achille Gaggia, assigned to Brevetti Gaggia SpA, Italian priority 23 March 1955, filed 9 March 1956, granted 11 August 1959. It is explicitly about "that kind of coffee which is known by the name Expresso". It is also the wrong machine — a later automatic design, not the 1947 lever — and it does not state nine bar.

Most telling: even the specialist literature declines to source it. The 2026 Physics of Fluids paper describes the standard as "often traced back to the force achievable in the first lever-operated espresso machines" and cites, for that, a food-history journal article and a popular reference book — not a patent, not a measurement. When a peer-reviewed paper hedges a historical claim to secondary sources, that is a good indication no primary source is available to be cited.

The commercial pages that assert it most confidently do not agree on why. One says plainly that historical records suggest Gaggia did not design the machine with a precise pressure in mind. Another concedes nobody has the exact answer and offers the average of a lever pressure curve as a guess. Under this site's sourcing rules a retailer history cannot carry a historical claim, so the honest statement is the unsatisfying one: nine bar is an inherited convention whose technical origin is undocumented in any primary source available here.

One physical detail in the story is independently sound and worth keeping. A spring-driven piston does not hold constant pressure; as the spring extends its force decays, producing a declining profile across the shot. So whatever nine bar described on a lever machine, it was a peak or an average of a curve rather than a setpoint. The pump era implemented it as a flat line held for the duration of the extraction, which is a different thing under the same name.

What the modelling work actually found

The most systematic published treatment is Cameron et al., Matter, 2020, which paired a mathematical model of extraction with experimental measurement.

Its central result concerns grind, not pressure. A model assuming homogeneous flow through the bed predicts extraction yield falling monotonically as grind coarsens. Experiment instead shows a peak, with lower yields at both very coarse and very fine settings. The authors attribute the fine-side collapse to flow ceasing to be homogeneous: small particles wedge into the gaps between larger ones, water oversamples some regions and misses others, and reproducibility degrades. The model's homogeneous and inhomogeneous flow scenarios diverge most at the finest setting tested, differing by 13.1% in extraction yield at grind setting 1.1 against 6.1% at 1.3 and 2.6% at 1.5 — the divergence itself being the evidence for inhomogeneity at fine grinds.

Now the part that requires care, because it is easy to overread.

The paper reports that their espresso machine — a San Remo Opera three group — set to nine bar clogged at fine grind settings, and that the pressure was reduced to six bar to circumvent this difficulty. Their experimental series then ran at six bar.

What that establishes is narrow: in this particular apparatus, at the fine settings the authors wanted to explore, nine bar was not workable, so they changed it in order to run the experiment. It is a methodological choice made to obtain data, not a comparative finding that nine bar produces worse coffee. The study did not test nine against six with everything else held constant and report an outcome.

Separately, the model does predict that extraction yield can be increased by grinding finer, using lower water pressure, and using less coffee. That is a model prediction, and it is the paper's own framing. The strongest defensible statement combining the two is that the published model points toward lower pressure being useful, and that the experimental programme was conducted at six bar rather than nine — not that nine bar has been shown to be wrong.

The paper's headline practical recommendation is about dose and grind rather than pressure: reduce dry mass and grind coarser. Its café validation moved from 20 g finely ground to 15 g coarser at Tailored Coffee Roasters over a year and 27,850 drinks, reporting comparable extraction yield, shot times near 14 seconds, and about $3,620 in reduced coffee cost.

One incidental finding is worth recording because it contradicts a durable piece of folklore. The authors standardised tamping at 98 N, and report exploring a range of tamp pressures without observing appreciable variation in shot time or extraction yield.

What follows

• Establish which pressure your gauge reports before trusting any number from it. A reading taken with no portafilter installed distinguishes a pump gauge from a brew gauge immediately.
• Expect puck inlet pressure to sit below the gauge reading during flow, and to vary through the shot as bed resistance changes.
• Treat blind-basket setting as valve calibration, not as measurement of what the coffee sees.
• Stop thinking of the puck as a fixed resistance. Above about five bar its permeability falls as you push harder, so additional pressure buys progressively less flow.
• Treat nine bar as an inherited convention whose technical origin is undocumented. That is a reason to hold it loosely, not evidence that it is wrong.
• If shots clog or scatter at fine settings, lower pressure is a reasonable thing to try. Two independent lines point that way: the published extraction model, and the poroelastic result showing the high-pressure returns are not there in the first place.

What was not checked

The exact grind settings at which Cameron et al. observed clogging are in Supplemental Information that the publisher blocks from automated access and that could not be found in an institutional copy. This article does not rely on them. The machine, the clogging at nine bar and the six-bar experimental setting are all stated in the main paper.

No manufacturer figure was found for the pressure loss across an E61 gicleur. The one service document checked in full — La Marzocco's 62-page GS/3 manual — specifies the pump setting and acceptable gauge range but never names the restriction between them. The familiar 1.5-bar offset traces instead to one forum post about one machine, so it is presented here as that rather than as a general constant.

The same limitation applies to the origin of nine bar. The 2026 peer-reviewed paper traces it to secondary literature rather than a primary document, while Gaggia's own history names a patent for its 1938 machine and none for the 1947 lever machine. Two well-motivated sources fail at the same point. That makes the origin undocumented; it does not prove that no primary record exists.

The Home-Barista references were checked against Internet Archive captures because the live site refuses automated requests. Usernames and post timestamps are therefore included in the references rather than replaced with unattributed “forum consensus”. For manufacturer manuals that assign several incompatible “normal” bands to one brew gauge, see the Synesso MVP, Strada AV, and Rocket reconciliations.