Shrinkage gets discussed as a property of materials. "ABS shrinks, PLA doesn't." That framing is why so many people compensate the wrong things and leave the right ones alone.
A material gives you a percentage. A percentage is not an error you can act on. Multiply it by the dimension you care about and you get millimetres, and only then can you compare it against the tolerance you actually need. The same spool is worth compensating on one feature of a part and not on another.
What a 120 mm dimension really measures
Every figure below is a 120 mm dimension printed at the top of that material's stored shrinkage range, with nothing done about it.
| Material | Shrinkage | Comes out at | Model it at |
|---|---|---|---|
| PLA-CF | 0.3% | 119.64 mm | 120.3611 mm |
| PLA | 0.5% | 119.4 mm | 120.603 mm |
| PETG | 0.6% | 119.28 mm | 120.7243 mm |
| ABS, ASA, polycarbonate | 0.8% | 119.04 mm | 120.9677 mm |
| Nylon 6, unfilled | 2% | 117.6 mm | 122.449 mm |
| Polypropylene | 2.5% | 117 mm | 123.0769 mm |
The spread from the best-behaved filament to the worst is nearly three millimetres on one dimension. One of those parts drops into its slot; the other needs a file taken to it.
Notice also that the correction is not the shrinkage. To recover 120 mm from a material that loses 2.5%, you model at 123.0769 mm — a scale factor of 102.5641%, not 102.5%. Small difference; it matters on long parts, and the compensation calculator does the division for you.
The same percentages on a 20 mm feature
Now shrink the part. A 20 mm boss in ABS at that same 0.8% comes out at 19.84 mm — sixteen hundredths under. In unfilled nylon at 2% it comes out at 19.6 mm.
For most purposes the ABS figure is nothing. It is inside the noise of first-layer squish and nozzle-width calibration, and compensating for it will not produce a measurably better part. The nylon figure is four tenths of a millimetre, which is the difference between a bearing that presses in and a bearing that falls out.
Same two materials, same two percentages, opposite conclusions — because the dimension changed. This is the whole point: compensate dimensions, not materials. Decide by asking whether the predicted error exceeds the fit you need, and what clearance a press fit needs is the other half of that comparison.
The stored range is the real problem
Here is the part that gets left out of most advice, and it is the reason to be careful rather than enthusiastic.
The site stores PLA at 0.2–0.5%, and that range is honest: pigment load, brand and cooling all move it. Suppose you compensate a 120 mm dimension for the top of that range, modelling at 120.603 mm, and then print it on a spool that actually shrinks at the bottom of the range. The part comes out at 120.3618 mm.
Compare that with doing nothing at all on the same spool: 119.76 mm.
Uncompensated you were about a quarter of a millimetre small. Compensated from the published maximum you are over a third of a millimetre large, which is worse, and worse in the direction that cannot be fixed with a file. That is not a quirk of PLA. Work it through for every material in the first table and the result is the same each time: compensating for the top of a stored range, on a spool that turns out to sit at the bottom of it, leaves the part further from nominal than doing nothing would have. Compensating from a range you did not measure is not a safe default. It is a bet, and it is a bet against the widest, least-documented variable in the material.
Compensate from a measurement, not from a table
The reliable procedure has one extra step and it removes the guesswork entirely.
- Print a calibration block in the actual spool you will use, with a dimension in the same size class as the one you care about.
- Measure it cold, and measure it in the axis you care about.
- Feed intended and measured into the calculator, which returns the scale factor rather than a percentage.
- Apply that factor to that axis, in that material, and write it down next to the spool.
Everything in the first table is a fallback for people who have not done this, exactly like a published spool weight is a fallback for people who have not weighed their reel. The shrinkage and tolerance guide covers the measurement side properly, and how to compensate for shrinkage is the condensed version.
Which axis you are compensating
Not one material on this site contracts equally in all three directions, and the reason is mechanical rather than chemical: the build plate holds the bottom of the part while it cools, so X and Y contract more freely than Z does. Uniform scaling in the slicer therefore over-corrects Z on most parts.
The calculator asks which axis for that reason. If you only ever compensate one, compensate the long horizontal one, because that is where the millimetres are.
The materials where this is not optional
Three groups genuinely change how you design rather than how you scale.
Unfilled nylon. PA6 is stored at 0.8–2.0%, the widest and largest range of any common filament, and it is why fibre-filled nylon exists at all — the fibre restrains contraction along the extrusion direction and cuts the figure to a quarter. How much nylon shrinks has the numbers; the practical answer is often to buy the filled grade.
Polypropylene. Crystallisation-driven, up to 2.5%, and it keeps moving for hours after the print finishes. A PP part measured straight off the plate is not the part you will have tomorrow.
Unmodified PET. Same mechanism. An amorphous polymer shrinks predictably; one that crystallises as it cools does not, and it can shift again if the part is later heated — which is also why annealing changes dimensions and why annealed parts need their own measurement rather than the annealing guide's general advice about strength.
Holes are a different problem, mostly
A hole in a uniformly contracting part scales down with everything else, so in principle the same factor fixes it.
In practice, undersized holes usually are not a shrinkage problem at all. Extrusion width, over-extrusion on the inner perimeter, and the first layer squishing outward all conspire to shrink circular features more than the material does, and none of them respond to a scale factor. If your holes are tight and your outside dimensions are right, the cause is in the extrusion, not the contraction — why holes print too small separates the two. Scaling the whole model to fix a hole will simply make everything else wrong, and designing the fit properly is the durable answer.
For the record on the two materials that started this: how much PLA shrinks and how much ABS shrinks both have the ranges, and the gap between them is smaller than their reputations suggest.