Why 3D Prints Are the Wrong Size: Shrinkage & Nozzle Flow

The model says 20.00mm. The calipers say 20.31mm. The 5mm hole you designed for a 5mm pin takes a 4.6mm drill bit and nothing else, the lid that should click onto the box has to be forced, and the bracket that measured perfectly at 30mm is 1.5mm long at 200mm. Dimensional error in 3D printing is not one fault with one fix — it is four separate mechanisms that happen to produce the same complaint, and each one leaves a different signature on the part in your hand. Read the signature and the fix takes an evening. Guess, and you will chase a flow number that was never the problem. This guide sorts them in the order they actually occur, and ends with the case where the right answer is a shop with calibrated machines.

This post contains affiliate links. If you purchase through these links, 3D Prototyping Hub may earn a small commission at no extra cost to you.

This post also contains Amazon affiliate links. As an Amazon Associate we earn from qualifying purchases.

Measure Before You Diagnose

Almost every wasted weekend on this problem starts with a description instead of a number. "Tight", "loose" and "a bit off" cannot be told apart; 20.31mm can.

Get a pair of digital calipers and take four measurements before changing anything: an outside dimension on a small feature, the same kind of dimension on something 150mm or longer, a hole, and the overall height. Let the part reach room temperature first — a bracket measured straight off a 60°C bed reads large and shrinks while you are writing the number down.

Those four numbers separate the causes on their own:

Over-Extrusion: Fat Walls Make Big Parts and Tight Holes

This is the most common cause and the most misread. The slicer plans a toolpath down the centre of each extruded line and assumes the line comes out a known width. If the real line is wider, half the surplus lands outside the intended surface. Outer dimensions grow, and inner surfaces — the walls of every hole and pocket — move inward by the same amount. Large outside, small inside, both by roughly the same fraction of a millimetre, is the fingerprint.

Three things cause it. The flow multiplier is set high, often because someone raised it to paper over a different problem. The filament diameter in the profile says 1.75mm while the spool is actually running 1.71mm, so the slicer under-estimates how much plastic each millimetre of feed delivers. Or the nozzle orifice has worn oversize, which is slow, invisible, and normal on abrasive filament — a 0.4mm brass nozzle that has printed a few kilos of carbon-fibre or glow-in-the-dark material is not 0.4mm any more. Keep spare 0.4mm nozzles in the drawer and treat them as consumables; our guide to nozzle sizes and materials covers which diameter belongs on which job.

Calibrate it properly rather than by feel. Print a single-wall open box with one perimeter, no top or bottom and no infill, measure the wall in several places, and set flow to desired width ÷ measured width × current flow. That is a direct measurement of what the machine actually deposits. If flow lands more than about five percent from where it started, something mechanical is wrong and you are compensating for it — the same trap the under-extrusion checklist describes from the other direction.

Holes Are Small on Purpose, and That Is Geometry

Undersized holes are the single most reported dimensional complaint, and on a correctly tuned machine they are not a fault at all.

Two effects stack. A circular hole is sliced into straight segments whose chords sit inside the true diameter, so the printed opening is inherently smaller than the drawing. Then the inner perimeter is laid against a wall that is already there, and the extruded plastic is compressed slightly inward rather than being free to spread. On a 0.4mm nozzle the two together commonly cost somewhere in the region of 0.1 to 0.4mm of diameter, worse on small holes because the faceting error grows as the hole gets smaller relative to the line width.

You have three honest fixes and one bad habit to avoid:

• Oversize the hole in CAD for the specific fit you need. Most reliable, because it travels with the model.
• Use the slicer's hole compensation (called XY hole compensation, hole horizontal expansion or similar). Correct here because the error is a fixed offset, not a percentage.
• Print undersize and drill or ream to size. The only method that gives a genuinely round hole, and the right call for a bearing seat or a dowel.
• Do not raise the flow rate to open holes up. It works for exactly that model and makes every outside dimension worse.

For anything that takes a screw, stop fighting the hole and change the interface: heat-set threaded inserts melted into a slightly-wrong hole give a metal thread at a known size, and they survive assembly cycles that a printed thread does not.

Shrinkage: the Error You Cannot Tune Away

Every thermoplastic leaves the nozzle molten and contracts as it cools to room temperature. That contraction is a percentage, which is why it hides on a calibration cube and ruins a long part. Typical published ranges:

Treat those as starting points, not specifications — the real figure depends on the grade, the chamber temperature and how much of the part is solid. Measure a 150mm or 200mm test bar in your own material, work out the percentage, and apply it as a scale factor, not as horizontal expansion. Scale fixes a percentage error at every size; a fixed offset does not. This is the same physics behind warping, so a material that is fighting you on flatness is usually the one costing you dimensions too.

Moisture makes the whole exercise unrepeatable. Wet filament extrudes inconsistently, so the same dimension varies between prints off one spool and no calibration number ever settles. Dry the spool first — a heated dryer that also feeds the printer keeps it dry through a long job, and our comparison of filament dry boxes covers the practical options. Choosing a material on its shrinkage and stability rather than on colour is what stops this recurring on the next project.

One Axis Wrong: Belts, Pulleys and Play

Flow and shrinkage errors do not care which direction a feature runs. So if X measures true and Y is short, the fault is in the motion system, and it is mechanical.

Work through it in this order. Belt tension first — a slack belt lets the carriage lag behind the motor at each direction change, and the part lands short in that axis while sharp corners pick up ringing. A plucked belt should give a clean low note rather than a dull slap, and a proper tensioner holds a setting that a thumbscrew and a spring will not. A belt whose teeth have rounded off will not hold tension at all and needs replacing.

Pulley grub screws next: one must sit on the flat of the motor shaft, and a pulley that creeps produces an error that changes every print. Then carriage play — rock the toolhead by hand, and if it moves, the V-wheels are flat-spotted or the eccentric nuts have loosened. Fresh wheels and eccentrics remove the slop at source; tightening a belt to hide it just loads a worn bearing.

Test on a large rectangle, not a cube. A 0.5 percent error is 0.1mm on 20mm and easy to dismiss, and 1mm on 200mm where you cannot miss it.

Height Is Its Own Problem

Z rarely fails for the same reasons as X and Y.

The first layer sets the datum. A nozzle squashed too close removes real material from the total height and flares the base outward — the elephant's foot that makes a part refuse to sit flat and read oversize at the bottom. Set the gap to a repeatable number with a feeler gauge rather than a sheet of paper, and use the slicer's elephant-foot compensation for the residual.

Height is quantised. A 10.05mm boss printed at 0.2mm layers becomes 10.0mm or 10.2mm, because there is no half layer. Design critical vertical dimensions as multiples of your layer height, or accept the rounding.

The Z screw can bind. A dry, gritty or bent lead screw skips a fraction of a turn under load and every layer above it stacks short. Clean it and re-grease with a PTFE lubricant; if the error is a consistent percentage instead, the fault is the Z steps-per-mm value and belongs with extruder calibration.

Fix It in This Order

1. Measure four features with calipers on a cooled part, and write the numbers down.
2. Check the filament diameter in the profile against five caliper readings along a metre of the actual spool.
3. Dry the spool if it has been open for weeks — otherwise nothing you measure next will repeat.
4. Calibrate flow from a single-wall test, and extruder steps before that.
5. Fix the mechanics — belts, pulleys, wheels, lead screw — before touching a compensation setting.
6. Apply a scale factor for material shrinkage, measured on a long part.
7. Apply hole or XY compensation last, for the fixed geometric offset that remains.

Where a machine has been rebuilt to its limit and the numbers still will not hold, the honest answer is newer hardware rather than another evening: both ELEGOO's FDM machines and motion spares and Anycubic's FDM range and parts list spares by model, which avoids matching a belt pitch or a wheel bore by eye.

When the Tolerance Decides the Method

A desktop machine that has been through the list above will hold a few tenths of a millimetre on most parts, most of the time. The phrase that matters is most of the time. Repeatability across materials, colours, seasons and batches is a different property from best-case accuracy, and it is the property industrial equipment is actually sold on.

So when a drawing calls out a bore, a bearing seat, a sealing face or a mating dimension that has to be right the first time — or when the part is large enough that a percentage error becomes millimetres — the shorter path is a provider whose machines are calibrated and inspected as routine. Our guide to tolerances and accuracy explains how to specify those dimensions so they come back correct. Then browse providers by location and process, send the same STEP file to two or three, and compare the quote against another fortnight of test cubes.

Hero photograph by ZMorph All-in-One 3D Printers on Unsplash.