Published shrinkage figures are ranges because shrinkage genuinely is one. Your spool, your bed temperature and your part's geometry land somewhere inside that range, and no table can tell you where. So the useful skill is not memorising percentages — it is running one measurement and converting it into a number your slicer can use. That takes about twenty minutes and it is good for as long as you keep buying the same filament.
What one calibration square actually tells you
The test piece is a solid square, printed flat, big enough that the error is larger than your calipers' repeatability. A 60 mm square works: at PETG's shrinkage a 20 mm cube moves by less than most people's measuring error, which is why cube-based calibration produces so much argument.
Printed in PETG on a 70 °C bed, mine measured 59.74 mm across X. Feed that in and the arithmetic falls out:
- Observed contraction: 0.4333%
- Scale factor to apply: 1.004352, which most slicers want typed as 100.4352%
- A nominal 60 mm feature should therefore be modelled at 60.2611 mm — a correction of 0.2611 mm
The shrinkage calculator wants the intended and the measured dimension, not a percentage, because the percentage is only an intermediate step and deriving it by hand is where the sign gets lost. The direction is worth stating plainly, because it trips people up: the part came out small, so the compensation makes the model bigger. You are not shrinking anything; you are pre-stretching it.
Why the same square gave a different answer in Z
The same part measured 59.94 mm in height — a contraction of 0.1%, roughly a quarter of what X did. That asymmetry is not measurement noise, and it is the reason this site stores an isotropy flag alongside every shrinkage range.
While a layer is cooling it is welded to the layer beneath it and, at the bottom, to a heated plate. It cannot pull inward freely. In X and Y each layer contracts across its whole width with nothing but the previous layer to restrain it; in Z the contraction of any one layer is a fraction of that layer's own thickness, and the stack is repeatedly re-heated from above as printing continues.
There is a second reason to distrust a Z scale factor, and the calculator warns about it: first-layer squish and Z-offset are a fixed error, not a proportional one. If your nozzle sits four hundredths of a millimetre low, every part is that much short in Z regardless of height. Scaling the model spreads a constant error proportionally, which makes tall parts worse while appearing to fix short ones. Correct Z-offset with Z-offset, and reserve scaling for X and Y.
The spread inside a published range is bigger than people expect
Take PETG's stored range of 0.2 to 0.6% and apply both ends to a 120 mm dimension. At the bottom of the range the part comes out at 119.76 mm; at the top, 119.28 mm. The compensated model dimension is 120.2405 mm in the first case and 120.7243 mm in the second.
Nearly half a millimetre separates two figures that are both correct for "PETG". If your part has a mating dimension across 120 mm, choosing a number off a table is choosing a coin flip. Measure.
Materials sort themselves into three practical groups by how much this matters:
- Ignore it unless the part is long or mates with something. PLA and its blends, at 0.2 to 0.5%, plus filled PLA where the fibre restrains contraction further.
- Compensate deliberately. PETG, ABS, ASA and HIPS all sit in the 0.2 to 0.8% band, where a 100 mm feature is visibly off.
- Compensate and expect to iterate. Unfilled nylon at 0.8 to 2.0%, polypropylene at 1.0 to 2.5%, and PEEK. These are also the materials where the number moves with chamber temperature, so a factor derived in summer may not hold in an unheated garage in winter.
Holes are a different problem wearing the same clothes
A hole in a printed part is almost always undersized, and people reach for the shrinkage factor to fix it. That is the wrong tool, and applying it introduces an error in every other dimension of the part.
Two effects dominate a hole's diameter, and neither is thermal contraction. The first is polygonal approximation: a circle is printed as a series of straight segments that lie inside the true circle, so the printed hole is inscribed rather than circumscribed. The second is extrusion width — the nozzle lays a finite-width bead whose inner edge bulges toward the centre on a tight radius.
Both effects scale with hole diameter differently from the way shrinkage does, which is why a single scale factor cannot fix holes and dimensions at the same time. The fix belongs in the model: oversize holes at design time, or use your slicer's hole-compensation setting, which exists precisely because this is a separate phenomenon. If holes are your actual complaint, holes printing undersized is the specific diagnosis, and designing tolerances and fits covers how much clearance a given fit needs.
When compensating makes things worse
Three cases, all of which I have watched people get wrong:
Assemblies printed as a set. If a shaft and its housing are both printed in the same material on the same machine, they shrink together and the fit is preserved. Compensating both makes them both bigger and changes nothing except your confidence. Compensate when the printed part mates with something not printed — a bearing, a bolt, a extruded aluminium profile.
Parts smaller than your measurement error. A caliper repeatable to two hundredths of a millimetre cannot resolve 0.4% across 20 mm. If your test piece is small, you are calibrating your calipers, not your printer.
After a material change of any kind. A scale factor is a property of a filament on a machine at a bed temperature, not of the printer. A different colour of the same product line can differ, because pigment loading changes both the density and the crystallisation behaviour of the polymer.
The procedure, so you can repeat it in twenty minutes
- Print a solid square, at least 60 mm on a side, flat on the plate, at the bed temperature you actually use.
- Let it cool to room temperature completely. A part measured warm is still moving, and the reading will be generous.
- Measure X and Y separately, at three positions each, and take the median rather than the mean — one bad reading skews a mean and does not skew a median.
- Put the intended and measured figures into the calculator and read the slicer scale.
- Apply it to X and Y only, and re-print the square once to confirm the correction landed.
Step five is the one people skip, and it is the only step that tells you whether any of the previous four were right.
Keep a note of the figure against the spool. When you finish that spool and buy the same product again, re-run the square rather than assuming — it takes twenty minutes and the alternative is discovering the difference in a part you have already glued.