The vertical walls came out clean. The gently domed top looks like a contour map — a dozen concentric ledges you can catch a fingernail on, each one a fraction of a millimetre high and a couple of millimetres wide. Stair-stepping on a curved surface is the one print defect that is not a defect: it is the direct, unavoidable consequence of building a smooth shape out of flat slices, and it appears on a perfectly calibrated machine printing a perfect file. What people get wrong is assuming it scales with layer height. It scales with layer height divided by the slope of the surface, and that second term is what turns an invisible 0.2mm ledge into a visible 2mm terrace. Once you can predict which surfaces will show it, the fix is usually free. When it cannot be fixed on the machine you own, the provider directory lists shops running processes that resolve it in hardware.
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.
Read the Surface Before You Touch the Layer Height
Four different faults get described as "rough on the curves", and only one of them responds to a thinner layer. Look at which way the roughness runs and where on the part it sits.
| What the surface looks like |
What it actually is |
Where to start |
| Horizontal terraces, worst where the surface is nearly flat, absent on vertical walls |
Stair-stepping — geometry |
Orientation, then adaptive layers |
| Flat vertical panels down a round feature, like a polygon |
Mesh faceting from the STL export |
Re-export with a tighter chord tolerance |
| Horizontal bands at a regular spacing including on vertical walls |
Z banding — mechanical or flow |
Z banding |
| Rough and drooping only on downward-facing slopes |
Unsupported overhang, cooling |
Overhangs sagging |
| One ridge running vertically up the whole curve |
Layer start point |
Visible seams |
The first two are the ones that get confused, because both read as "faceted". The test takes a second: stair-stepping runs around the part, following the layers. Faceting runs up it. If you can see flat panels on a cylinder when you look at it from above, your mesh is the problem and no printer setting will help.
The Arithmetic That Decides How Bad It Will Be
This is the whole article in one line, and it is worth doing in your head before you slice.
Each layer is a flat slice of constant thickness. Where the surface slopes, each slice sticks out past the one below it by a horizontal distance — the step width — and that distance is the layer height divided by the tangent of the surface angle measured from horizontal:
step width = layer height ÷ tan(slope angle)
At a 0.2mm layer height, that works out as:
| Surface angle from horizontal |
Step width at 0.2mm layers |
How it reads |
| 90° (vertical wall) |
0 |
Perfect |
| 45° |
0.20mm |
Invisible |
| 30° |
0.35mm |
Fine on a matte surface |
| 20° |
0.55mm |
Noticeable |
| 10° |
1.13mm |
Obvious terracing |
| 5° |
2.29mm |
Contour map |
| 2° |
5.73mm |
Looks like a wedding cake |
The non-obvious consequence: a sphere is smooth at its equator and terraced at its poles, from the same layers. So is every fillet, every rounded lid, every domed canopy. The problem is concentrated in the small part of the surface that lies nearly flat, which is also why cranking the layer height down globally is such a poor trade — you pay for it over the whole part to fix a few square centimetres.
Measure a step with a pair of digital calipers before you accept any of this. If the measured ledge matches the arithmetic, the machine is behaving and you have a geometry decision to make. If it is much larger, or it appears on vertical walls too, you are looking at one of the other three rows in the first table.
Orientation Is Free and It Beats Everything Else
Rotating the model is the only change that attacks the denominator rather than the numerator, and it costs nothing.
A surface at 5 degrees from horizontal is the worst case. The same surface rotated to 45 degrees produces a step a tenth as wide — the geometry has not changed, only its relationship to the build direction. On a part with one critical cosmetic face, tilting the whole model 30 or 40 degrees and accepting supports elsewhere is almost always the better deal than halving the layer height across the entire print.
Three things to weigh when you rotate:
- The surface you move out of the shallow zone has to be replaced by something. Rotating a lid so the dome is steep usually puts a flat rim into the shallow zone instead, which is fine, because a flat face parallel to the bed has no steps at all. Aim the shallow zone at surfaces that are either perfectly flat or hidden.
- Supports mark the surface they touch. Trading stepping for support scars on the same face is not a win. Point the critical face upward and let the supports land on the back. Removal damage is a separate subject covered in why supports will not come off.
- Strength changes with orientation too. Layer adhesion is the weak axis, so a bracket rotated for cosmetics may break in the direction it is now loaded — see why prints snap along layer lines before rotating a functional part.
If the part cannot be rotated, say because it is a tall vase with a shallow shoulder at the top, that is the point at which settings and hardware become the answer rather than the first resort.
Layer Height, Adaptive Layers and the Nozzle Underneath Them
Halving the layer height halves every step width and roughly doubles the time. That is an honest trade and sometimes the right one, but it is rarely the efficient one.
Adaptive or variable layer height is what most slicers now offer instead, and it is built exactly for this geometry. The slicer measures the local slope and assigns thin layers where the surface is shallow and full-height layers where it is steep or vertical. On a typical domed part that recovers most of the visible benefit for a small fraction of the time cost, because the shallow zone is a minority of the height.
The floor on all of this is the nozzle. A 0.4mm nozzle meters a layer reliably somewhere between roughly 0.1mm and 0.3mm; push much below that and the extruder cannot control such a small volume per millimetre of travel, so you trade terracing for inconsistent flow — which produces under-extrusion rather than a smoother curve. If you genuinely need finer layers, change the nozzle: a 0.2mm nozzle makes 0.05mm layers practical and cuts every step width on the part by a factor of four against 0.2mm layers. It also slows the print roughly in proportion to the reduction in extrusion width, clogs more readily, and demands clean filament. The trade-offs per diameter are laid out in our guide to nozzle sizes.
One setting that is often mistaken for a fix: ironing. It smooths top solid surfaces that lie flat, by running the hot nozzle back across them, and it does nothing at all for a sloped surface, because there is no flat top layer there to iron.
Faceting Comes From the File, Not the Printer
If the roughness runs vertically — flat panels down the side of a hole or a cylinder — the mesh is coarse and the printer is faithfully reproducing it.
An STL stores curves as flat triangles. The exporter decides how many, using a chord or deviation tolerance, and a loose default turns a 20mm hole into a visible twelve-sided prism. The fix is upstream: re-export from CAD with a tighter tolerance, or step up to a format that carries the real geometry. STEP keeps the mathematical surfaces rather than an approximation of them, which is why a shop will usually ask for it — the trade-offs are in STL versus STEP for 3D printing. For models you downloaded rather than made, the faceting is already baked into the file and remeshing is the only route. Design-side habits that avoid the whole class of problem are collected in our design for 3D printing guidelines.
Material and Colour Decide How Much You See
The steps do not change size with the filament. How visible they are changes enormously.
A glossy surface produces a specular highlight, and on a terraced curve that highlight traces every single step edge as a bright line. A matte surface scatters the same light diffusely and the edges disappear into the form. The practical effect is large enough that the identical model in glossy black and in matte PLA reads as two different qualities of print. Dark saturated colours are the worst case because the only thing your eye has to work with is the highlight; mid-tone matte greys are the most forgiving, which is why prototype shops print so much of it.
This is cosmetic rather than dimensional, and worth being clear-eyed about. If the part has to measure right, the steps are still there and a caliper still finds them — that is dimensional accuracy, a different problem with different causes.
When Resin Is the Honest Answer
If curved surfaces are the entire point of the object — a figure, a jewellery pattern, a display model, a scale body shell — the process is the lever rather than the settings.
An MSLA machine builds in 0.01 to 0.05mm layers, against 0.1 to 0.3mm on a desktop extruder. On that same 5-degree surface, a 0.02mm layer leaves a step around 0.23mm wide instead of 2.29mm, which is below the threshold at which most people read a surface as terraced. In-plane resolution comes from the LCD pixel pitch rather than from an extrusion width, so small features survive that an 0.4mm nozzle simply cannot form. ELEGOO's resin printer range and Anycubic's US store both publish the two numbers that matter here — minimum layer height and XY pixel pitch — alongside the vats and cure equipment; the comparison across models is in our resin printer round-up, and opaque grey model resin is the sane default for anything where you are judging a form.
Resin is not a free upgrade. It brings a wash and cure routine, gloves, ventilation and a consumable screen and film, and it has its own shallow-surface artefact — exposure variation across a near-horizontal face produces horizontal lines on resin prints that look superficially similar and have a completely different cause. The process comparison, including where each one stops making sense, is in SLA versus FDM.
Fixing the Part You Already Printed
Finishing works well on stair-stepping, and there is one rule that decides whether the curve survives it.
Fill the valleys; do not grind down the peaks. Taking abrasive to bare plastic removes the step edges and leaves the recesses, so a hand-sanded dome comes out subtly flattened — and on a tight fillet, faceted in a new way. Spray a high-build filler primer first and let it do the filling, then sand the primer back. Filler primer with a set of sanding sticks is the standard kit for this, and the backing matters as much as the grit: flexible foam-backed sticks and needle files follow a curve where a rigid block puts a flat spot in the middle of it.
Work up through the grits with an assortment from 120 to 400, wet from 400 upward to stop the paper loading, and re-prime whenever bare plastic shows through. Two or three cycles will take a visibly terraced dome to something that photographs smooth. For organic shapes where sanding would soften detail you want to keep, a self-levelling clear epoxy coat fills the steps without removing anything — at the cost of a few tenths of a millimetre of added thickness, so keep it off mating surfaces. Adhesion is the usual failure here rather than the filling: scuff and degrease before primer, or you get the outcome described in why paint peels off 3D prints. The full consumables list is in our finishing and painting supplies guide.
When to Stop and Hand It Over
Three situations make this someone else's machine rather than a longer evening with yours.
A cosmetic surface on a one-off is rarely worth the hours — hand-finishing a terraced body shell is skilled, repetitive work and a finishing shop does it faster and better. A batch is worse, because hand-finishing does not scale and will not be consistent across twenty parts. And a shallow curve that also has to hold a tolerance is the case where neither thin layers nor filler primer helps, since one is too slow and the other adds material. That is industrial SLA, DLP or machining territory. The providers in our directory are listed by location and process, and many list post-processing and finishing as a service in its own right, which is worth asking about even when the printing happens in your own workshop.
Run the order and most parts resolve in one reslice: find the shallowest surface that has to look good, check its angle against the table above, rotate the part if you can, switch on adaptive layer height if you cannot, and only then reach for a smaller nozzle or a different process. The arithmetic tells you which of those will actually pay before you spend eight hours finding out.
Hero photo by Efe Yağız Soysal via Unsplash.
