Laser Kerf & Press-Fit Calculator
Why an uncompensated tab-and-slot joint is loose by twice the kerf, not once — and why the sheet usually varies more than the beam does.
What to draw
Measuring the kerf: one square against a row
An uncompensated joint is loose by twice the kerf, not once
The beam is centred on the drawn path and removes half a kerf either side of it. So an outer feature loses a full kerf across its width, and an inner feature gains one. A tab is outer and its slot is inner — so the two errors add. Draw both at 3.00 mm with a 0.2 mm kerf and you cut a 2.80 mm tab into a 3.20 mm slot: 0.40 mm of slop where you wanted none. That's why finger-jointed boxes rattle, and why the fix feels twice as large as people expect.
"Offset by half the kerf" and "the part comes out a full kerf small" are both true, which is why the advice sounds contradictory. Half is the offset per side; a full kerf is the error per dimension, because a width has two sides. Neither is the compensation for a joint. And note that a single global compensation setting — one offset applied to everything — corrects the size of your parts and leaves the joint exactly as loose as it was, because the slot and the tab need offsets in opposite directions.
Measure the kerf on a row of squares, not one. Cutting ten squares in a strip and butting them together puts your ±0.05 mm caliper error against a 2.00 mm shortfall — ±2.5%. One square puts the same error against a 0.20 mm shortfall, which is ±25%: you'd be measuring your calipers rather than the beam. Same tool, ten times the precision, one extra minute of cutting. (It buys precision, not accuracy — a mis-zeroed caliper stays wrong.)
And the sheet varies more than the kerf does. Nominal 3 mm birch ply runs 2.6–3.1 mm in practice — a 0.5 mm spread, 2.5× a typical kerf, and it varies within one sheet as well as between them. Cut a slot for the nominal 3 mm and a sheet at the far end is out by 0.4 mm, which swamps the compensation you just calculated. MDF is the exception at 0.2 mm of spread, which is why it's worth prototyping in. Measure the sheet in front of you.
- Kerf is a property of the cut, not the material. It moves with power, speed, focus, lens, air assist and how dirty the optics are. A figure from someone else's machine is barely a starting point.
- The cut isn't a rectangle through the thickness. The beam has a waist, so it's narrowest at the focus and wider above and below. Focusing at mid-thickness roughly halves the taper — which is why a joint can be snug from one face and loose from the other.
- Acrylic doesn't do press fits. It's brittle and doesn't compress, so interference becomes hoop stress and shows up as crazing weeks later rather than a crack you'd notice at assembly. Aim for a slip fit and glue it.
- Cut a test joint before the whole sheet. One tab and one slot costs a minute and settles every number on this page for the material actually in the machine.
How to use
- Measure the kerf on a row of squares, not a single one.
- Measure the sheet you are about to cut, in several places.
- Draw the slot narrow by a kerf and the tab wide by one.
- Cut one test joint before committing the whole sheet.
Frequently asked questions
How much should I compensate for laser kerf?
For a joint, by twice the kerf — which is more than most guides say. The beam removes half a kerf either side of the drawn line, so an outer feature loses a full kerf across its width while an inner feature gains one. A tab is outer and its slot is inner, so the two errors add. Draw both at 3 mm with a 0.2 mm kerf and you cut a 2.80 mm tab into a 3.20 mm slot: 0.40 mm of slop.
Is kerf compensation half the kerf or the full kerf?
Both figures are correct, which is why the advice sounds contradictory. Half the kerf is the offset per side. A full kerf is the error per dimension, because a width has two sides. Neither of them is the compensation for a joint, which is twice the kerf because the slot and the tab move in opposite directions.
Why do my finger-jointed boxes rattle?
Almost always because the kerf was compensated once rather than twice, or not at all. The tab comes out a full kerf narrower than drawn and the slot a full kerf wider, so a joint drawn with zero clearance assembles with twice the kerf of play. At a typical 0.2 mm kerf that is 0.4 mm, which is plainly visible and quite enough to make a box feel cheap.
Can I just use my software global kerf offset?
Not for joints. A single offset applied to every path corrects the outside dimensions of your parts, which is useful, but it moves the slot and the tab the same direction — so the joint ends up exactly as loose as it started. Joints need the slot drawn narrow and the tab drawn wide, which is two different offsets rather than one.
How do I measure my laser kerf accurately?
Cut a row of squares in a strip, butt them together and compare the total against the number of squares times the drawn width. The shortfall is one kerf per square, so ten squares gives ten kerfs of shortfall and divides your reading error by ten. One square puts a 0.05 mm caliper error against a 0.20 mm shortfall, which is 25 per cent; ten squares makes it 2.5. Same calipers, one extra minute.
Does measuring more squares make the result more accurate?
It makes it more precise rather than more accurate, and the distinction matters. Averaging over ten squares divides your random reading error by ten, but it does nothing about a systematic problem — a mis-zeroed caliper or a habit of squeezing too hard stays exactly as wrong at any number of squares. Zero the calipers first, then use the strip.
What is a typical laser kerf?
Around 0.15 to 0.3 mm on a CO2 machine through thin sheet, but the figure is a property of the cut rather than of the material. It moves with power, speed, focus height, lens, air assist and how dirty the optics are, so a number from someone else machine is barely a starting point. Measure it on yours, and re-measure whenever anything on that list changes.
Why does my press fit work on one board and not another?
Because the sheet varies more than the kerf does, and nobody budgets for it. Nominal 3 mm birch plywood runs about 2.6 to 3.1 mm in practice, a 0.5 mm spread that is two and a half times a typical kerf, and it varies within a single sheet as well as between them. A slot cut for the nominal thickness will be tight on one board and loose on the next through no fault of the settings.
Which sheet material is most consistent for press fits?
MDF, by a clear margin — around 0.2 mm of spread on nominal 3 mm, against 0.5 for birch plywood. That is why it is worth prototyping a design in MDF even when the finished piece will be plywood: the fit you dial in is repeatable, so you are testing your geometry rather than the lumber. Acrylic sits in between at roughly 0.4 mm.
Should acrylic parts be a press fit?
No. Acrylic is brittle and does not compress, so interference goes into hoop stress rather than into the joint, and it crazes or cracks weeks later rather than at assembly — which makes the cause hard to spot. Aim for a slip fit with about 0.1 mm of clearance and let solvent cement do the holding. Plywood and MDF are the materials that genuinely hold a press fit.
Why is my joint tight on one face and loose on the other?
That is kerf taper. The beam has a waist, so the cut is narrowest at the focus and wider above and below it — the cut is a shallow hourglass rather than a rectangle. Focusing at mid-thickness rather than at the top surface roughly halves the effect, which is why mid-thickness is the usual default on anything over a couple of millimetres. Flipping a part over sometimes fixes a fit for the same reason.
How much clearance should a laser-cut joint have?
About zero for a glued box in plywood, since you want it snug enough to hold itself square while the glue sets and loose enough to assemble by hand. Around 0.05 mm of interference for a press fit meant to hold without glue in plywood or MDF. Roughly 0.1 mm of clearance for anything you will take apart, for acrylic, and for assemblies that have to go together in a particular order.
🔒 This tool runs entirely in your browser. Nothing you enter is uploaded, logged, or stored.