Annealing means holding a finished part above its glass transition, but below its melting point, long enough for the polymer chains to rearrange, then cooling it slowly. For PLA this is a genuine and useful process. It is also the subject of some of the most confident and least supported claims in desktop printing, so this page grades them.
What is happening in the oven
PLA can crystallise, but when it is printed it barely does. The plastic goes from melt to solid in seconds, and the chains freeze in a disordered, largely amorphous arrangement before they have time to fold into crystals.
Hold the part above its glass transition and the chains regain enough mobility to organise. Crystalline regions nucleate and grow, and those regions do not soften at the glass transition — they hold their structure until the crystals themselves melt, far higher up.
That single mechanism explains everything annealing reliably does, and it also explains the side effects.
Graded by how well it is supported
Well established: heat resistance rises, substantially. This is the effect worth having. An annealed PLA part holds its shape well above the temperature at which the same part would sag untreated, because the crystalline fraction is no longer governed by the glass transition. This is repeatable, it is mechanistically understood, and it is why the process exists at all.
Well established: the part changes size. Crystalline PLA is denser than amorphous PLA, so a part that crystallises contracts. Reported figures vary widely with grade and schedule, and the movement is not the same in every axis, since the printed part is anisotropic before it goes in.
Well established: stiffness rises. Annealed parts are noticeably more rigid.
Mixed evidence: tensile strength. Published results go both ways. Some studies show modest gains, some show losses, and the differences track the grade of PLA, the temperature, the hold time and the cooling rate rather than pointing to one answer. Anyone quoting a single percentage improvement is quoting one experiment.
Reasonably well supported, and awkward: impact toughness falls. A more crystalline polymer is generally more brittle. Parts that need to survive being dropped are often worse after annealing, which sits directly against the popular framing of annealing as a strengthening process.
Weakly supported: better layer adhesion. Plausible — chains have mobility to diffuse across interfaces — and not well characterised for FDM parts in practice.
Not supported: that annealing turns PLA into an engineering material. It raises PLA's ceiling. It does not make PLA into ABS, and a part that needs to be tough and hot is still a job for a different polymer.
The dimensional problem, with numbers
This is the practical obstacle and the reason annealing is rarer than the enthusiasm for it suggests.
Take a 100 mm dimension. At a reported contraction of 1.5%, the annealed part measures 98.5 mm, and to land on size you would have to model it at 101.5228 mm — a correction of 1.5228 mm applied before printing. At the low end, 0.5%, it measures 99.5 mm and needs a 0.5025 mm correction. The spread between those two corrections is the real problem. Both 1.5% and 0.5% are plausible outcomes for the same part in the same oven, and until you have annealed one and measured it, the shrinkage calculator has nothing better than your guess to work from.
A whole millimetre separates those two compensations across 100 mm. That is the difference between a press fit and a rattle, and the spread is not something you can look up — it depends on your filament's grade, your schedule and your part's geometry. The compensation is also not the same in every axis, so a single scale factor will get one dimension right and another wrong.
The consequence: annealing is for parts whose dimensions do not have to be exact. A vase, a bracket with clearance holes, a housing that bolts on with slots. Anything with a fit is a job for calibration first, and the shrinkage and tolerance method applies twice over.
A schedule that works, and the ways it goes wrong
- Temperature. Above the glass transition, well below the melting point. Somewhere in the 80 to 110 °C region for PLA. Hotter is faster and increases the risk of the part slumping under its own weight.
- Support the part. Sand, salt, plaster or a bed of glass beads. A part annealed unsupported in an oven can sag visibly, and thin unsupported features will.
- Hold time. Tens of minutes rather than hours for a small part; longer for a thick one, since it is the core that has to reach temperature.
- Cool slowly. Cooling in the switched-off oven with the door shut is the standard method. Rapid cooling locks in stress and can crack the part.
- Expect the surface to change. Annealed parts often lose gloss and can develop a slightly frosted look as crystallites scatter light.
Domestic ovens overshoot their setpoint badly on the way up, so use an oven thermometer and let the oven stabilise before the part goes in. A part put into an oven still climbing to temperature is the usual cause of a slumped result.
Materials other than PLA
The picture changes entirely, so do not generalise from the above.
PETG and ABS are amorphous and do not crystallise, so annealing them relieves internal stress rather than raising heat resistance — sometimes useful for dimensional stability, not a route to a hotter part. Nylons and PET do crystallise and can be annealed with real gains, but they also move dimensionally when they do, which is precisely why unmodified PET is so unpredictable in the first place — see PET against PETG.
Is it worth doing
Anneal when the part is heat-limited, geometrically forgiving, and would otherwise have to be reprinted in a harder material on a machine you may not own. That is a genuine and useful niche.
Do not anneal to make a part stronger. The evidence does not support it, the impact behaviour probably gets worse, and the hour spent would be better spent adding walls or turning the part on the plate.