---
title: "You cannot buy a good square anymore"
author: "Hal Mireles (@rustbelt)"
date: 2026-04-30T08:33:38.256Z
updated: 2026-04-30T08:33:38.256Z
canonical: "https://jot.place/@rustbelt/you-cannot-buy-a-good-square-anymore"
description: "Test the square you already own by reversal, then hold it against what DIN 875 actually allows. Eight squares, my own measurements, one of them firewood."
tags:
  - "craft"
  - "machining"
  - "measurement"
  - "metrology"
  - "tools"
---

# You cannot buy a good square anymore

Three squares came into my garage this year in blister packs with a certificate in the box. All three certificates said DIN 875/0. One of the three actually was.

The certificate is not exactly a lie. It is a claim about a model, printed by the ten thousand, and nobody put an indicator on the square in your hand. Ask for a serial number and a recorded deviation for that unit and watch what happens.

There are real reasons the cheap ones go off. The blade gets ground and then hardened, and heat treat moves steel, which is why the good makers harden first, grind after and lap last. The inside corner is not relieved, so a burr down in the corner holds the work off and the square reads open inside and tight outside. And blade straightness is a separate error nobody advertises: a blade can be bowed 5 tenths along its length and still pass a squareness check taken at one point.

:::info
Reversal works because the thing you are measuring changes sign when you turn it around and the thing you are measuring with does not. Subtract the two readings and you get the part. Add them and you get your instrument. That is the most useful idea in shop metrology and it costs nothing to use.
:::

## Against a straight edge, with a scriber

1. Get a reference edge you trust, and prove it first. Scribe a line along the edge, flip the edge end for end, scribe again along the same line. If the edge is bowed, the two lines cross in the middle and separate at the ends, and the separation is twice the bow.
2. Clamp the reference down. Set the square's beam against it with the blade flat and scribe along the blade. One pass, sharp carbide, fine line.
3. Mirror the square, so the same blade face looks the other way while the beam still rides the same edge, and bring the blade edge onto your line at the near end.
4. Look at the far end. Any gap, or any crossing, is twice the error of the square.

Steady divergence end to end is squareness error. Lines that cross in the middle mean the blade is bowed, and now you know two things. But with a loupe and good light you can only see about a thousandth of separation between a scribed line and a blade edge, so this test tells you your square is junk and it will never tell you your square is good.

## Against an indicator, with a number at the end

You need a surface plate, an indicator reading in tenths, and a stand with a vertical slide. The slide does not have to be perpendicular to anything, which people find hard to believe until they do it once.

1. Clean the plate and stone the square. Stand it beam down, blade up, one blade face toward the indicator.
2. Touch off a quarter inch above the beam, zero, then run the slide up to a quarter inch below the tip. Write the reading down with its sign, call it R1, and note the swept length.
3. Do not touch the stand. Rotate the square 180 degrees about a vertical axis so the same blade face points the other way, and repeat the sweep. Call it R2.
4. Error of the square over the swept length is (R1 minus R2) divided by 2. Whatever is wrong with your column sits in both readings with the same sign, so it drops out.

Equal readings with the same sign mean your square is good and your column leans. Equal and opposite means all of it is the square. A grade B plate is 3 tenths overall, which sounds too coarse for this, but the beam never leaves one four inch patch and inside that patch the plate is far better than its overall number.

To compare against a grade you need an angle, not a length. With e the deviation and L the swept length:

$$\theta \approx \frac{e}{L} \text{ radians} \qquad \theta_{\text{arcsec}} \approx 206265 \cdot \frac{e}{L}$$

Four tenths over 4.000 inches is 0.0001 radians, about 21 seconds of arc. DIN 875 works in microns against the length of the shorter leg,[^1] and its four grades come out as 3, 7, 15 and 30 microns at 100 mm, near enough a 4 inch blade, which in inches is 0.00012, 0.00028, 0.00059 and 0.00118. So a 4 inch square with 4 tenths of error is a grade 1 square carrying a grade 0 certificate. Not a scandal. A factor of two, in the direction that suits the seller.

## What I have actually measured

All mine, by the reversal above, three runs each, worst run reported. Sweep was 2.500 inches on the block, 3.500 on the 4 inch squares, 5.500 on the 6 inch and 8.000 on the long blades, so these figures run a little kind.

| Square | Blade | Sold as | Error | Microns | Earns |
|---|---|---|---|---|---|
| Starrett 20-4, my father's, bought 1974 | 4 in | hardened master square | 0.0001 | 2.5 | 00 |
| Starrett 55 combination head | 12 in | combination square | 0.0026 | 66 | outside grade 2, and that is fine |
| Import solid square, unit one | 4 in | DIN 875/0 | 0.0006 | 15 | 1, right on the line |
| Same model, unit two, same box, same day | 4 in | DIN 875/0 | 0.0011 | 28 | 2, and outside it scaled to the full leg |
| Import double square | 6 in | "precision ground" | 0.0004 | 10 | 1 |
| Ground 1-2-3 block, American, 1980s | 3 in | not sold as a square at all | 0.0001 | 2.5 | 00 |
| Cast iron try square from my grandfather's chest | 8 in | woodworking | 0.006 | 150 | firewood |
| Cylindrical square I ground and lapped myself | 6 in | mine | 0.0002 | 5 | 0 |

The combination square is in that table on purpose. It is not defective. It is a layout tool with a moving joint and a clamp, and half the arguments about squares are somebody holding a combination head against a hardened master and getting upset about the result.

The two identical imports out of the same box, differing by a factor of two, tell you what the paper in the box is worth. Unit two read R1 = +0.0016 and R2 = -0.0006, which works out as 0.0011 inch of squareness error and 0.0005 inch of lean in my own stand. The tools are not bad for the money. The number on the certificate has nothing to do with either of them.

## What to buy instead

Old American hardened squares. A Starrett 20 series, a Brown and Sharpe, a Lufkin out of somebody's grandfather's chest, costs about what the import costs new. Test it the day it lands and send it back if it fails. Granite is the other good answer, because it does not rust and will not take a burr, and used granite is cheap because it is heavy and awkward to ship.

Or make one. Grind a piece of two inch round to length between centers, lap the ends on the plate, prove it by reversal, and you have a cylindrical square better than anything you were about to buy. Mine took a Saturday. If you do buy import, buy two of the same and keep the better one. That is not cynicism, that is what the unit to unit spread is.

Then two habits. Stone the beam and blade faces before every use, because at least half the squares people bring me to check are perfectly good squares wearing a burr from a drop. And hang them on the wall. A square that lands on a concrete floor is a layout tool now, and it will not tell you it changed jobs.

Everybody in the trade knows the certificate means nothing and everybody keeps buying the box with the certificate in it, which is a particular arrangement people make with themselves. @bindery/justified-text-on-the-web-is-a-lie-we-agreed-to is about a different one and it is the best thing I read this year. And @dust/limiting-magnitude-from-first-principles is the clearest writing I have seen out of any trade on what an instrument can actually resolve.

[^1]: DIN 875 sets permitted deviation from perpendicularity against the length of the shorter leg, in microns: grade 00 is 2 plus L/100, grade 0 is 5 plus L/50, grade 1 is 10 plus L/20, grade 2 is 20 plus L/10, with L in millimetres. I measure over a sweep shorter than the leg, so my figures flatter the square a little compared with how the standard would judge it.