What Actually Stops a Bullet — And Why Most Published Numbers Are Wrong

The problem is not bad data. It is data with the wrong caption.

If you search for how deep a bullet goes into pine, concrete, or water, you will find numbers. They circulate. They get quoted in forums, then in articles, then in other articles that cite the articles. Almost none of them cite a measurement.

We went looking for the measurements. What we found was a specific, repeating failure — and it is not that people lie. It is that a number gets separated from its definition, and after that it is indistinguishable from a measurement.

Six patterns kept appearing. Each one is illustrated below with the source that exposed it.

Pattern 1: The unit is not what it says

A widely-quoted table gives penetration into pine boards for thirty handgun loads, in inches. A 9mm 147gr Hydra-Shok reads "9 in." A .223 55gr JSP reads "6 in."

Those are not inches. They are board counts.

The proof is that the same author described the same test in prose on a different site. The comparison is exact:

Five for five. And the structure held twelve panels total — so "12" is a ceiling, not a measurement. Nine of the thirty loads read exactly "12 in." They did not stop at twelve inches. They ran out of target.

Pattern 2: The material is special, sold as ordinary

Concrete penetration figures for handgun rounds trace back, more often than not, to SACON® — and SACON is not concrete in the sense you mean.

From the patent (US 9,121,675, US Army ERDC): SACON is "a low-leaching, foamed concrete" — "low-density, fiber-reinforced" — and its purpose is stated plainly: "the cellular structure of this type of concrete permits incoming bullets and fragments to bury themselves in the concrete without producing ricochets."

It is a bullet trap. It is installed at ranges — West Point, Fort Knox, Fort Bragg — specifically because it catches bullets instead of deflecting them. The patent even notes that penetration depth "can be adjusted by modifying the SACON density."

The reported figures (Hudson et al., 1999) are "1 inch (25 mm) for a .38 caliber pistol to 2.55 inches (63 mm) for 5.56 mm." Those are real. They are also for a foamed, low-density, purpose-built absorber — not for the reinforced structural concrete in a wall.

The same failure appears elsewhere:

• Clear Ballistics vs 10% ordnance gel. The lab that produced the most-cited handgun gel data says it outright: "We could mix up some FBI ordnance gel… but that still would not give us any data that we can directly compare." Their blocks are synthetic. The FBI protocol calls for 10% calibrated gelatin at 4°C. These are not interchangeable, and the source knows it.
• Armour steel vs mild steel. A 2023 study of "bullet splash" against steel is frequently cited for steel penetration. It tested Creusabro® 8000 and Durostat® 400 — grades whose stated application is "ballistic shields and barriers" — against .308 Winchester. Not A36 mild steel. Not 5.56.

Pattern 3: The method measures something else entirely

Three examples, all from real, competent research:

UL 752 is quoted for bullet-resistant glazing depth. It does not measure depth. It is pass/fail: "a product fails if penetration occurs." Level 1 means three shots of 9mm 124gr FMCJ at 1175–1293 fps must not get through. That tells you a bound, not a number.

Depth-of-penetration (DOP) testing is quoted for ceramic. It measures the backing, not the ceramic: "the ceramic tile is bonded to a backing material of semi-infinite thickness, and the penetration depth of the projectile gives a measure of the performance of the ceramic." The number you read is how deep the round went into the polycarbonate block behind the tile. It ranks armour. It does not describe the ceramic.

And sometimes the source refuses. The same handgun test that produced the board-count table also shot brick and cinder block. Its authors wrote: "we didn't record penetration data for either of these because it wasn't measurable like the wallboard and pine boards were. However, we did collect our visual impressions."

That is honest research. The numbers that later appeared for those materials did not come from it.

Pattern 4: A lower bound quoted as a measurement

"The round went through all four walls" is not a penetration depth. It is a floor.

A well-known drywall test reports XM-193 going through four ½-inch walls spaced ten feet apart. That establishes more than four walls — with no upper number, because the test ran out of walls.

A separate test with twenty-four ½-inch panels spaced 3.5 inches apart gives the number the first one could not: "terribly penetrative (19 panels, or nine walls)." And it defines the measurement explicitly: "'Penetration' is the number of sheets of half-inch sheetrock that the rounds fully penetrated; this number can be divided by two to get the number of 'walls'."

Nineteen panels × 0.5″ = 9.5 inches of gypsum. That is a measurement. "Went through four walls" is not.

Pattern 5: The famous number has no primary source

Ask what a 12-gauge slug does in ballistic gel and you will be told 18 to 20 inches. It is everywhere.

Martin Fackler measured 14. One-ounce (437gr) Foster slug, 1513 ft/s, 10% ordnance gel, bare — 14 inches, roughly a 1-inch wound channel. His method is documented: "shooting projectiles into 10% gelatin blocks kept at a temperature of 4 degrees C. Blocks (25 × 25 × 50 cm) were placed end to end."

Where does 18–20 come from? Fr. Frog, citing Fackler's own profile, gives the honest version: "Observed penetration for Foster type rifled slugs ranges from 14″ to around 18″ depending on brand and hardness." Fourteen is the measured point. Eighteen is the top of a range for harder slugs. Twenty appears in secondary sources with no measurement attached at all.

Pattern 6: The field simply never measured it

This one is not a failure of citation. It is a hole in the literature, and it is startling.

In 2017, researchers at the Netherlands Defence Academy published the only modern controlled study of bullet penetration into wood: six species, three 9mm projectiles, X-ray measurement of depth, five shots per velocity, velocities chronographed at 5 m and 18 m.

In their own literature review, they wrote: "Unfortunately, we did not find more recent scientific work in open source literature in which projectile penetration into wooden targets was studied."

The previous study they could find was by Metz — around 1838.

That is a 190-year gap. If you have ever wondered why the numbers for "how deep does a bullet go into a wooden door" are all vague — that is why. Almost nobody has measured it.

What the primary sources actually say

Here is what we found that is a measurement, with the source.

Concrete: two inches stops a rifle round — at any range

US Army FM 3-06.11, Table 7-3, "Penetration capabilities of a single 7.62-mm (ball) round," in inches:

Two inches of concrete, at every range tested. Not because the manual is imprecise — because concrete under confinement resists on a completely different scale than a handbook compressive strength suggests.

And penetration goes UP with distance

Look at the pine column again: 13 → 18 → 41 inches as the range increases. That is not a typo. The manual explains it directly:

This inverts the intuition that faster means deeper. A high-velocity rifle round arriving at close range is under such stress on impact that it destabilises and tumbles — and a tumbling bullet presents its side, not its point. It stops sooner.

Almost every penetration calculator — including ours — models velocity and ignores this. Ours reads a flat 16.8″ / 16.9″ / 17.0″ across those same three ranges. The manual says 13 / 18 / 41. We are not wrong by a scale factor. We are missing a mechanism.

Water: about one foot, and it is not the drag that stops it

The Naval Ordnance Laboratory shot rifle rounds into water in 1969 with a Fastax high-speed camera and pressure transducers (NOLTR 70-174, DTIC AD0713445, public release). Their conclusion:

M-16 ball at 991 m/s: effective slant range ≈ 12 inches. M-14 ball at 899 m/s: ≈ 18 inches, and that is the maximum.

Their theory section computes what a stable flat-nosed projectile would do: 11 feet for 5.56, 15 feet for 7.62. The measured result is one foot. The difference — a factor of eleven — is the cost of tumbling and break-up.

Their camera caught the moment: "a bubble-like cavity forming at the point where the bullet tumbled… apparently the bullet, by tumbling, has dissipated most of its energy at that point, causing the explosive-like bubble." The M-14 round came out in two major pieces.

They then spent the report engineering a projectile that would work underwater. Filing a flat nose onto a standard bullet failed immediately: "ineffective due to material failure, i.e., mushrooming of the end of the bullet." They needed a steel insert in a copper cup to get a straight track — and even the final design reached only 7.5 feet, not the theoretical 11–15.

Oak: the one wood number that exists

From the 2017 study, for European oak the preferred model is Robins-Euler (constant deceleration; the inertial term fits to essentially zero), with a dynamic resistance of 5,200 N. For a 9mm Ball projectile that gives:

Note the range: 240 to 370 m/s. Anything quoted outside it is extrapolation, whatever it is labelled.

The study also found something useful about hollow points in wood, without setting out to: "projectiles did not show deformation after penetration into Douglas, Pine and Oak targets." Combined with the well-documented observation that a hollow cavity "just fills up with pine" and the round fails to expand at all — in wood, a JHP behaves like an FMJ. Two independent findings, same conclusion.

What this means for our own numbers

We are not going to pretend this audit left us clean. It did not.

Against the sources above, our engine hits about half the time — and the misses are not random. Every one of them is too deep, and every one is a case where the bullet tumbles, fragments, or splashes.

The pattern is clean enough to state as a rule: the model is reliable for handgun rounds at 250–400 m/s and unreliable for rifle rounds at 800–900 m/s. Below the tumbling threshold it computes a stable projectile pushing through a medium, which is what actually happens. Above it, the round is coming apart, and we are still modelling it as a solid.

Concrete is our worst case, and four independent routes agree on it: Forrestal's empirical equation, the confinement multiplier implied by his S constant, the SACON figures, and FM 3-06.11. All four say the same thing.

Why publish this

Because the alternative is showing a number with a ±15% margin next to it when we know it is 3.4× off, and hoping nobody checks.

Every ballistics calculator produces numbers. Very few will tell you which ones they stand behind. We would rather be the one that says: here is the measurement, here is our model, here is the gap, here is the document.

If you want to check any figure in this article, the sources are below. Every one is public. Most took ten minutes to find and read — which is the uncomfortable part, because the numbers they contradict have been circulating for years.

Methodology notes

• What counts as a source. A measurement needs three things: an impact velocity from a chronograph, a described target (material, thickness, configuration), and a measured depth. Missing any one, it is not an anchor. "Went through" is a bound. "Visual impressions" is not a number.
• Check the range. A figure attributed to a study must fall inside that study's measured range. We caught one number quoted at 412.7 m/s from a study whose oak experiments ran 240–370 m/s — it cannot be a raw data point from that work.
• Watch for round-after-conversion. 412.7 m/s is 1354 ft/s. A suspiciously precise metric figure that becomes a plausible imperial one has usually travelled through a conversion, and sometimes through a mistake.
• Quantities must match. NOLTR reports "effective slant range" — where velocity drops to 300 fps, the threshold for a lethal wound per US Army Medical Service, 1962. That is not "where it stopped." Comparing it to a model's stopping depth is comparing two different things. When we recomputed our model at the same 300 fps threshold, a 2.7× discrepancy became 1.4×.
• Material variability is a floor, not an excuse. Peer-reviewed steel penetration testing reports ±12% shot-to-shot. A ceramic round-robin across six laboratories, same protocol, same ammunition, found the between-shot scatter so high that "the required number of tests is much higher than what is practically feasible." No model beats the material.

Related Analysis