Every functional print ends with the same quiet question, and most makers answer it the same lazy way: they look at the part, turn it over once, and decide it is fine. Then they press it onto the thing it was supposed to fit, and it is not fine. Too tight here, rattling there, one hole that never lined up in the first place.
This article is about closing that gap. Not with better slicer profiles, and not with another round of tolerance guessing, but with actual measurement — the unglamorous discipline of knowing, in numbers, whether the part in your hand matches the part in your CAD file. We will cover what your instruments can and cannot resolve, why a caliper can read perfectly while the part is still wrong, how to build fixtures that give you a clear yes or no, and what to record so the next print is better than this one.
The payoff is not aesthetic. It is the difference between a workflow that converges on a working part and one that keeps reprinting the same mistake with slightly different numbers.
Why You Are Probably Not Measuring, And Why That Costs You
Most makers own a caliper. Very few use it as an instrument. The typical pattern is that the caliper comes out after something fails, gets applied to the most convenient outside surface, and produces a number that either confirms the suspicion or is quietly ignored if it does not.
That is not inspection, it is reassurance. Real inspection has three properties that casual checking lacks.
- A defined target. You know the nominal dimension and the acceptable range before you measure, not after.
- An instrument that adequately resolves the feature. A tool whose uncertainty is a large fraction of your tolerance is not a measurement, it is a guess with decimals.
- A record. A number written down, with the direction it was off and by how much, becomes a correction you can apply. A number you forgot becomes nothing.
Skip any one of those and you are back to looking at the part and hoping. The irony is that inspection takes about ninety seconds per part once you have the habits, and it saves entire reprints. If you have ever used a systematic approach to dimensional accuracy to dial in your printer, inspection is what proves the dial-in actually took.
The Instruments, And What Each One Can Actually Resolve
There is a hierarchy here, and the important insight is that resolution and accuracy are not the same thing. A tool can display three decimals and still be useless for your task.
| Instrument | Typical resolution | Realistic accuracy on printed parts | What it is good for |
|---|---|---|---|
| Steel rule | 0.5 mm | ±0.5 mm | Overall envelope, obvious gross errors, sanity checks |
| Digital caliper | 0.01 mm | ±0.05 to 0.10 mm | Outside dimensions, depths, hole diameters with care |
| Micrometer | 0.001 mm | ±0.005 mm | Filament diameter, thin flat sections, wall thickness coupons |
| Feeler gauges | 0.05 mm steps | ±0.02 mm | Gap and clearance checks, sliding fit verification |
| Pin gauges | 0.01 mm steps | ±0.005 mm | Hole diameter verification, fit confirmation |
| Go / no-go gauge | By design | As accurate as you print it | Repeatable pass-fail on production parts |
Two things follow from that table. First, the caliper is the workhorse, but treat any reading below 0.05 mm as an indication rather than a fact. Second, the cheapest reliable instrument in the list is the pin gauge, and the most underrated is the feeler gauge set — both cost less than a spool and answer questions that calipers answer badly.
Where Calipers Lie On Printed Parts
Calipers are made for machined metal. Printed plastic violates their assumptions in several specific ways.
Layer lines turn a flat face into a staircase
A printed vertical wall is not a flat surface, it is a stack of bead-shaped ridges. Where the caliper jaw lands between two ridges it reads undersized; where it lands on a crest it reads oversized. On a 0.2 mm layer height the peak-to-valley variation of the surface alone can be 0.05 to 0.1 mm, which is often the same order as the tolerance you are trying to hold.
The fix is to measure in a way that is insensitive to where the jaw lands. On an outside dimension, measure across the full width and take the maximum of several readings rather than one. On a critical flat interface, consider that the surface you care about may not be the surface the caliper is touching.
Elephant foot makes the bottom of the part wider than the part
The first few layers squash outward. Measure at the very bottom and you will find your part is oversized by 0.1 to 0.3 mm, which will send you chasing a calibration error that does not exist. Always measure at least 5 mm above the build plate unless the bottom face is the functional one.
Soft materials deform under the jaws
Closing a caliper on a 95A TPU part compresses it. The number on the screen is partly a measurement and partly a squeeze report. For flexible materials, use light jaw pressure, or better, measure a gauge block alongside the part to see how much the jaws themselves flex.
Holes read small, and the error is directional
A printed hole is smaller than its CAD diameter, typically by 0.1 to 0.2 mm, and a caliper cannot measure it at all — the jaws measure across a chord, not a diameter, the moment you are off centre. Holes need pin gauges. This is not a nicety; if you are printing pivots, shafts and bushings, the hole diameter is the single number that decides whether the joint works.
The Four Measurement Errors That Matter
Before blaming your printer, rule out your technique. These four account for the overwhelming majority of bogus readings.
- Zero drift. Close the jaws, check the display reads zero, and do it every session. A caliper that reads 0.03 mm when closed will add 0.03 mm to every dimension you take all day.
- Jaw angle. Tilting the caliper by ten degrees on a 20 mm feature introduces roughly 0.3 mm of error. Feel the jaws seat flat against the surface before reading.
- Burrs and strings. A blob of stringing caught between jaw and workpiece reads as extra material. Brush the part and check the contact faces.
- Temperature. PLA and PETG move roughly 0.05 to 0.1 percent per degree Celsius. A part measured straight off a warm bed is not the part you will measure tomorrow. Let it cool, or note the difference.
Only after all four are ruled out is the remaining error attributable to the machine. If your part was showing shrinkage behaviour that does not match your expectations, check these first — measurement error and shrinkage look identical when you are only looking at one number twice.
Build Your Own Go / No-Go Gauge
Once a design is settled and you are producing more than one, the question changes from “what size is this?” to “is this one acceptable?” Those are different questions and they want different tools.
A go / no-go gauge is the standard answer. It is a printed block with two features: a go feature sized at the low limit of acceptability, and a no-go feature at the high limit. If the part fits the go and does not fit the no-go, it is in spec. No numbers, no interpretation.
| Feature | Go side | No-go side | Reading |
|---|---|---|---|
| Shaft diameter 10 mm, tolerance ±0.15 | Hole 9.85 mm | Hole 10.15 mm | Shaft must enter go, must not enter no-go |
| Slot width 4 mm, tolerance +0.20 / −0.10 | Pin 3.90 mm | Pin 4.20 mm | Pin must enter go slot, must not enter no-go slot |
| Wall thickness 2 mm, min 1.80 | Gap 1.80 mm | Gap 2.20 mm | Wall must not enter 1.80 gap, must enter 2.20 gap |
| Assembly clearance, 0.3 mm nominal | Blade 0.25 mm | Blade 0.35 mm | Blade must pass at 0.25, must not at 0.35 |
Three rules make the gauge trustworthy. Print it in the same material and orientation as the parts, because you are testing the process, not just the geometry. Print it with the features on the build plane wherever possible, since a gauge printed on its side has a different effective hole size than one printed flat. And label it — a gauge without legible markings becomes an unidentifiable lump within a month.
You can hold the gauge within roughly 0.05 mm if you print it flat with a dimensionally predictable PETG or a carbon-fibre reinforced PETG that is far stiffer than the base polymer and holds its shape better over time. A gauge made of a material that creeps will slowly lose its meaning.
Measuring The Things Calipers Cannot Reach
Some of the most important properties of a functional part are not dimensions at all, and they need different methods.
Wall thickness and shell integrity
Calipers measure outside dimensions. They cannot tell you whether the wall is solid or hiding an internal void. Print a test coupon with the same shell settings, break it, and measure the actual wall with a micrometer. If your slicer promised four perimeters and the wall measures 1.55 mm instead of the expected 1.68 mm, you have a flow or line-width discrepancy that will affect every part you make. This connects directly to how shell count and wall thickness carry load, because a wall that is thinner than designed is a wall that fails earlier than designed.
Clearance and running fit
For anything that moves, the number that matters is the gap, not the diameters. Feeler gauges answer this directly. Assemble the joint, then insert progressively thicker blades into the gap. If your design called for 0.3 mm of clearance and the 0.4 mm blade slides in easily, the joint is looser than intended and will rattle; if the 0.2 mm blade will not enter, it is tighter than intended and will bind. This is the fastest way to verify sliding surfaces that need to run without welding shut.
Flatness and seating
A part can be dimensionally perfect and still rock on a flat surface. Put it on a machined plate or a piece of glass and look for light underneath, or try to slip a 0.05 mm feeler gauge under the corners. A part that rocks will not seal, will not sit square, and will stress whatever it is bolted to. This matters enormously for anything that has to hold pressure across a gasketed face.
Squareness and alignment
Measure across both diagonals of a rectangular feature. If they differ by more than about 0.2 mm, the feature is not square. This catches skew that outside dimensions will happily hide, and it is the check that explains why an assembly that measures correctly still will not go together.
A Repeatability Test You Can Run In Twenty Minutes
Before you trust any measurement-driven correction, establish how repeatable your printer actually is. This is the single most useful diagnostic in this article, because it tells you whether the remaining error is correctable at all.
- Print five copies of a simple 20 × 20 × 10 mm block, all in the same job, same position on the plate.
- Let them cool completely — at least thirty minutes.
- Measure each one in the same place, three times, and record the maximum reading.
- Calculate the spread between the largest and smallest part.
| Spread across five identical parts | What it means | What to do |
|---|---|---|
| Under 0.05 mm | Process is highly repeatable; remaining error is correctable by calibration | Apply a dimensional correction with confidence |
| 0.05 – 0.15 mm | Normal desktop FDM variation | Correct only if your tolerance is wider than the spread |
| 0.15 – 0.30 mm | Something is loose or inconsistent | Check belts, eccentric nuts, spool drag, and filament feed consistency |
| Over 0.30 mm | Machine fault, not a tuning problem | Stop calibrating and find the mechanical cause first |
The lesson in that table is the one most makers learn late: if your spread is 0.25 mm, chasing a 0.05 mm calibration correction is wasted effort. Fix the repeatability, then calibrate. A part-scale answer to this is also worth knowing — if the same design fails the same way three times, the design may be loading a joint the way that makes it fail months later, and no amount of measurement will fix that.
Measurement By Material: What Changes
Material choice changes what you can measure reliably and how long a measurement stays true.
| Material | Dimensional stability after printing | Measurement notes |
|---|---|---|
| PLA and PLA+ high speed | Good short term, poor long term under load | Measure after cooling; numbers may drift over weeks if loaded |
| Matte PLA and silk PLA | Good, but surface texture complicates contact | Matt and silk surfaces vary more; use maximum of repeated readings |
| PETG and matte PETG | Good; slightly more creep than PLA | The default recommendation for gauges and fixtures |
| ABS, ASA and their CF / ASA-CF variants | Good once cooled; significant shrinkage during print | Measure at least an hour later; account for chamber temperature |
| PA6-CF20 | Excellent stiffness, but hygroscopic | Dry before printing or dimensions will not repeat |
| TPU 95A and high-speed TPU | Elastic; returns to shape after handling | Measure with minimal jaw pressure; repeatability is inherently lower |
| Tough resin / ABS-like resin | Very smooth surfaces, near-zero layer texture | Calipers behave well; watch shrinkage from cure |
Two material notes worth stating plainly. First, moisture is a dimensional variable, not just a print-quality one. Wet filament foams slightly at the nozzle, which changes extruded volume and therefore dimensions. If your parts are inconsistent, dry the spool in a heated drying cabinet or a sealed drying box before measuring anything. Second, fibre-filled materials wear nozzles, and a worn nozzle changes effective orifice size, which changes extrusion width, which changes dimensions. If you print abrasives, either run a hardened tungsten carbide nozzle or a spare nozzle on rotation and replace on a schedule rather than on failure.
Checking Fit Without Breaking The Part
Fit verification is where inspection earns its keep, and the mistake is to verify by forcing parts together. Press-on verification destroys the evidence — a joint that went together with a hammer tells you nothing about the joint, only about the hammer. For anything meant to come apart again, this is covered thoroughly in snap fits, press fits and inserts that open without breaking, but the measurement side has its own rules.
- Dry-fit before final assembly. If it needs force, measure before applying it — once deformed, you cannot tell whether the error was in the part or in the assembly.
- Check the interface, not the part. The caliper reading on a shaft is less informative than the feeler gauge reading in the bore it enters.
- Test the tightest feature first. Whichever dimension has the smallest allowance will fail first. Find it and measure there.
- Recheck after thermal cycling. A joint that fits cold may bind warm. Warm the assembly and measure again if it will see service above room temperature.
For threaded interfaces, the check is engagement depth rather than diameter. Measure how far a bolt or insert actually threads in, and compare that to the design intent. A thread that engages two turns short will hold far less load than one at full depth, and the external diameter will look completely normal — see thread forms, engagement depth and inserts for the numbers that matter.
What To Record, And Why It Compounds
Inspection only becomes powerful once it accumulates. A single measurement is an anecdote; twenty measurements of the same feature across different prints is a correction factor with evidence behind it.
Keep the record minimal, or you will stop keeping it. Five columns are enough.
| Field | Example | Why it earns its place |
|---|---|---|
| Part and feature | Bracket, bore diameter | Ties the number to something you can act on |
| Nominal and measured | 10.00 / 9.86 mm | The error itself, not just the reading |
| Direction | Undersized by 0.14 | Direction is what tells you how to correct |
| Material and spool | PETG, spool 3, dried 4 h | Catches material-side causes |
| Action taken | Raised XY hole compensation to 0.20 | Closes the loop so you do not repeat it |
After a few dozen entries you will see a pattern forming: a consistent directional offset that is a calibration fix, a random scatter that is a machine issue, or a material-specific bias that is a drying problem. Each of those has a different remedy, and only the record distinguishes them.
Inspection For Structural Parts
Dimensional checks confirm shape. They say nothing about whether a part will carry its load. For structural work you need a second, qualitative inspection that looks for the failure modes that matter.
- Look at the layer lines where stress is concentrated. Local inter-layer bonding is the weak link, and a feature that puts load across layers will fail there first.
- Flex the part deliberately and listen. Crackling or ticking indicates internal separation that no external dimension will reveal.
- Check whether load runs through a sharp transition. If it does, the part needs ribs, gussets and fillets rather than tighter tolerances.
- Verify the load path is continuous. A wall that is thick everywhere except at one transition is a wall that fails at that transition. Mounts that crack at the screw hole are the classic example.
- Confirm the part is stiff enough, not merely strong enough. A part that deflects under load may be within every dimensional tolerance and still useless, which is what designing sections for stiffness is about.
One structural check is worth doing on anything critical: measure the part, load it to service load, unload it, and measure again. A permanent set of more than about 0.1 mm means the part yielded, and a part that has yielded once is already weaker. This is the practical early warning for the slow failure modes described in why functional printed parts fail months later.
Checklist Before You Trust A Print
- Calipers zeroed at the start of the session, not assumed.
- Part cooled to room temperature — at least thirty minutes for anything substantial.
- Measured at least 5 mm above the build plate to avoid elephant foot.
- Maximum of at least three readings taken on every critical dimension.
- Holes verified with pin gauges, not calipers.
- Gaps and clearances verified with feeler gauges.
- Both diagonals measured on any rectangular feature.
- Flatness and rock checked on a machined surface.
- Filament dried if the material is hygroscopic, and nozzle condition confirmed if it is abrasive.
- Every number recorded with direction and any correction applied.
Where To Start, If You Start Somewhere
If you take one thing from this article, take the repeatability test. Twenty minutes, five small blocks, one number at the end. It tells you whether your remaining error is something you can calibrate away or something you have to repair, and it will save you from the most common wasted effort in this hobby: carefully correcting a printer that is not consistent enough for correction to mean anything.
If you take two things, add the go / no-go gauge. It converts a measurement problem into a decision, which is what you actually want when a part is in your hand and a machine is waiting.
And if you take three, start writing the numbers down. Measurement you do not record is measurement you will do again next month, to no better result. That habit is what separates a workflow that converges from one that simply repeats.