A hexagon table prints two numbers that look independent. One is s, the width across flats — the figure stamped on a spanner. The other is e, the distance across the corners — what a caliper reads when you turn the part thirty degrees. They are not independent. A regular hexagon's corner-to-corner diagonal is exactly 1.154701 times its width across flats, because the circle through the corners has a radius equal to the side length, and the side equals the flats. Set one and the other follows.
That turns two columns into a test. 484 rows in this database print both: 454 hexagon bolts and screws across 12 threads, 30 hexagon nuts across 15. The bolts come from four standards — DIN 933, ISO 4017, DIN 931 and ISO 4014 — and the nuts from two, DIN 934 and ISO 4032. If the two figures described one hexagon every time, they would divide out to the same number on all of them.
The ceiling has one occupant
1.154701 is a maximum, not a target. It is reached only when the real width across flats equals the nominal one, with nothing taken off for tolerance. On 26 of the 27 sizes the diagonal is below that, and the ratio e/s runs from 1.0900 up to the ceiling.
One size sits above it: the M2 nut, printed at 4 mm across flats and 4.62 mm across corners. 4 × 1.154701 is 4.6188, and 4.6188 rounded to two places is 4.62. That size clears the ceiling by 0.0012 mm, which is a rounding step rather than a hexagon. On that one size the diagonal was multiplied out from the flats rather than measured on a part.
That exception is worth knowing precisely because it is the only place in the data where the two columns are one number written twice. On the other 26 sizes they are two.
What the corner implies
Nothing in either table labels the last column below. It is a division: the diagonal read back through 1.154701 to the width across flats it is consistent with at its narrowest point. It is a floor for the real hexagon, not a second nominal size.
| Thread | Flats s (mm) | Bolt e (mm) | Nut e (mm) | Difference (mm) | Corner implies (mm) |
|---|---|---|---|---|---|
| M2 | 4 | — | 4.62 | — | 4.001 |
| M2.5 | 5 | — | 5.45 | — | 4.720 |
| M3 | 5.5 | — | 6.01 | — | 5.205 |
| M4 | 7 | 7.66 | 7.66 | 0 | 6.634 |
| M5 | 8 | 8.79 | 8.79 | 0 | 7.612 |
| M6 | 10 | 11.05 | 11.05 | 0 | 9.570 |
| M8 | 13 | 14.38 | 14.38 | 0 | 12.453 |
| M10 | 16 | 17.77 | 17.77 | 0 | 15.389 |
| M12 | 18 | 20.03 | 20.03 | 0 | 17.346 |
| M14 | 21 | 23.35 | 23.35 | 0 | 20.222 |
| M16 | 24 | 26.75 | 26.75 | 0 | 23.166 |
| M18 | 27 | 30.14 | 29.56 | 0.58 | 26.102 |
| M20 | 30 | 33.53 | 32.95 | 0.58 | 29.038 |
| M22 | 34 | 37.72 | 37.29 | 0.43 | 32.666 |
| M24 | 36 | 39.98 | 39.55 | 0.43 | 34.624 |
Read the last column against the second. The M2.5 nut, on 5 mm flats, is consistent with a hexagon measuring 4.720 mm across the flats at its tightest — 0.28 mm under nominal. At the other end the M20 bolt head, on 30 mm, is consistent with 29.038 mm, 0.962 mm under.
The millimetres lost grow with size. The fraction lost does not: it runs from 94.40% of nominal on the M2.5 nut to 96.79% on the M20 bolt. The spanner is made to the nominal width and the hexagon is made smaller than it, which is the whole reason a spanner is a sliding fit on a new part and a hammer on a worn one. Nothing in either column says so; the two numbers only say it together.
The nut and the bolt head split at M16
On the 12 threads both families carry, the width across flats is identical — one spanner fits either. The corners do not agree over the whole range. On 8 of those threads the two diagonals are the same to the last printed digit; on the other 4 the nut's is smaller: 0.43 mm on M22 and M24, 0.58 mm on M18 and M20.
M16 is the last thread where they agree. At M18 the nut prints 29.56 mm across corners while a bolt head prints 30.14 mm — same 27 mm flats, two hexagons.
The nut's own ratio shows the break better than the diagonal does. It runs from 1.0900 at M2.5 up to 1.1146 at M16, then falls to 1.0948 at M18 — a drop of 0.0198 on one step, which lands it within 0.0005 of the M4 nut. Over the same step the bolt head's ratio keeps climbing, from 1.1146 to 1.1163.
The two parts that a single spanner size serves are therefore not the same hexagon above M16. Anything that has to pass over both — a socket, a box-end, a clearance pocket in a fixture — is sized by the larger of the pair, and the nut is the smaller one.
The bolt ladder breaks somewhere else
The bolts have their own break, at a different thread. The ratio climbs from 1.0943 at M4 to 1.1177 at M20, then falls to 1.1094 at M22 — a step of 0.0083, the largest fall in the bolt ladder, and not at the thread where the two families part.
The cause sits in the flats column, not the corners. At M22 the width across flats is 34 mm, which is 1.545 times the thread rather than the 1.5 that holds either side of it. The spanner ladder has no 33 mm size to hand the M22 thread, so it takes the next rung up — and the same diagonal under a wider spanner is a smaller ratio.
That is the useful reading of the column. e/s is not a property of the thread and not a property of the standard; it is what the flats ladder left behind on that one size.
What to do with two numbers
- Take the spanner size from the flats column and the clearance from the corners column. They answer different questions and the second is usually the one that stops a socket fitting.
- Above M16, a nut and a bolt head that take the same spanner are not the same shape. If one tool has to handle both, size it to the bolt head and accept the play on the nut.
- Do not read
eas the size. It is a minimum across corners, and on every row here except the M2 nut it sits under what the nominal flats would give. The gap between the two is the tolerances, not a second standard.
Where this stops
- The tolerance band. These are one nominal flats figure and one minimum diagonal. Neither says where a particular part sits inside the band, and a plated or worn hexagon is outside both.
- Twelve-point sockets. They grip corners rather than flats, and a width-across-flats figure does not describe the shape they engage.
- Internal hexagons. On a socket-head screw
sis a hole, not an outside width, and the 1.1547 relation does not apply; that column is a different measurement under the same letter. - Why the families diverge. The data shows where the diagonals part and by how much. The tolerance classes that would explain the split are not in it, and this page does not guess at them.
- What a worn tool does. Neither column says anything about what happens to a corner after a few hundred tightening cycles.