The printer probes thirty-six points before every job, the mesh renders as a tidy little landscape on the screen, and the first layer is still translucent in the back-left corner. Nothing is broken. Auto bed levelling measures the shape of your bed and compensates for it — it does not level anything, and it does not set the height your nozzle prints at. Those two sentences account for most of the disappointment, and once you separate the jobs the fix is usually cheap and often free.
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 I earn from qualifying purchases.
What the Probe Is Actually Doing
The sequence is the same on every machine that has one. The probe touches a grid of points, the firmware stores a height at each, it interpolates a surface between them, and while the first few millimetres print it moves Z up and down to follow that surface. Then, on most firmware, the correction fades out and the machine goes back to printing in a flat plane.
That is a narrow job, and here is what sits outside it:
| What people expect it to fix |
What it actually does |
| The nozzle-to-bed distance |
Nothing — that is the Z-offset, and you set it |
| A bed that is mechanically tilted |
Compensates within range; the tilt is still there |
| A dent or crease in the sheet |
Averages it away, because it is smaller than the mesh spacing |
| Plastic not sticking |
Nothing — adhesion is surface, temperature and cleanliness |
| Drift over the first hour |
Nothing, if the mesh was probed cold |
| An inconsistent result print to print |
Nothing, if the probe itself is not repeatable |
Read down the right-hand column and the pattern is clear. Levelling gives the firmware a shape. Almost everything that goes wrong with a first layer is a height, a surface or a measurement, and no mesh has ever had an opinion about any of those.
The Z-Offset Is a Separate Number, and It Is Yours
The probe triggers at the probe's height. The plastic comes out of the nozzle. The gap between those two things is the Z-offset, and nothing in the levelling routine discovers it for you.
This is the one to check first, because its signature is unmistakable: the first layer is wrong the same way everywhere on the plate. Uniformly translucent and poorly bonded means the nozzle is too high. Uniformly glassy, ridged, with plastic squeezing out sideways means too low — and that is also where elephant's foot comes from. A uniform fault is never a levelling fault, because levelling only ever produces differences across the plate.
Set it against something with a known thickness. A feeler gauge is a measured 0.05mm or 0.10mm; a folded sheet of paper is somewhere between 0.08mm and 0.12mm depending on the paper, which is why the same technique gives two people two different machines. Then confirm it the only way that counts: print a single-layer patch, let it cool, and measure it with calipers against the layer height you asked the slicer for.
Also remember what invalidates it. A new nozzle, a new build sheet, a re-seated probe, a different probe mount — each one moves the offset, and each one is routinely fitted without re-setting it. That is a large share of the "it got worse after I upgraded it" cases in why 3D printer upgrades don't fix print quality.
A Mesh Is an Interpolation of the Points You Touched
A 3×3 mesh on a 220mm bed samples nine points about 110mm apart and draws straight lines between them. If your plate dishes gently in the middle, the mesh knows the middle is low only if it happened to probe there, and everywhere else it is guessing in straight lines across a curve.
Three consequences follow, and all three are testable tonight:
- Density matters more than people think. Going from 3×3 to 5×5 or 7×7 roughly halves the distance the firmware has to guess across. Most firmware supports it; re-probe after you change it, and re-probe after you move the machine.
- The probe cannot reach the edge. It is mounted offset from the nozzle in X and Y, so the outermost rows of the mesh are extrapolated rather than measured. Parts placed right at the plate edge are printing onto the firmware's best guess. This is also why batches behave differently from single parts — see why 3D prints fail when you print a full plate.
- A flatter plate needs less of the mesh. This is the honest argument for borosilicate glass: not that it prints better, but that there is less shape for a coarse mesh to miss. The trade is grip, and the full surface comparison is in our build plate guide.
The Probe May Not Be Measuring the Surface You Print On
Each probe type has a specific blind spot, and knowing yours saves an evening.
Inductive and capacitive probes sense metal — which means an inductive probe reads the steel sheet underneath your PEI, not the PEI surface the plastic actually lands on. A dented sheet, a variable-thickness coating, or a crumb of filament trapped under the magnetic mat all put a real bump in the print surface that the probe cannot see. Capacitive sensors additionally drift with temperature and humidity.
Deploying probes of the BLTouch kind add a mechanical pin that has to seat identically every time. A loose mount, a tired spring or a cable tugging the carriage turns the mesh into noise.
Nozzle-tap and strain-gauge systems probe with the nozzle itself, which removes the offset problem entirely and introduces a new one: a blob of residual plastic on the tip changes the trigger height. Wipe the nozzle hot before every probing cycle, every time.
The test that settles it costs nothing. Probe one single point ten times and compare the readings. A spread above roughly 0.02mm means you have a mechanical problem, not a levelling problem, and the mesh you have been carefully regenerating was describing the probe rather than the bed. A dial indicator on a magnetic base gives you the same number independently, and the same setup run across the plate tells you how flat the bed is — which is the measurement nobody has and everybody guesses at.
Cold Mesh, Hot Machine
Probing a cold bed and then printing on a hot one is measuring one machine and operating a different one. An aluminium bed expands as it heats, the spring steel on top of it expands at a different rate, and the frame carrying the gantry moves as the machine and any enclosure warm through. The numbers are small and the first layer is a small-numbers problem.
Heat the bed and hotend to printing temperatures, hold them for at least ten minutes, then probe. If your first print of the session is consistently better or worse than the ones after it, this is your cause and the soak is your fix. A bed insulation pad shortens the soak and narrows the gap between the centre of the plate and its colder perimeter. The same thermal story explains a good deal of why the same file prints differently on two printers.
The Bed Still Has to Be Mechanically Flat
Compensation has a range, and working at the edge of it means the Z axis is making constant corrections while trying to lay a smooth line. Tram the bed by hand first — with the probe's own mesh as your guide, since it is already telling you which corner is low — and let the mesh handle only the remainder.
If the bed drifts back out every few weeks, the springs are the reason. They compress unevenly and relax over time, which is why a bed set in March is out by June. Silicone levelling columns do not creep in the same way and are a genuine fix rather than a tuning step. Check the stud length suits your bed before ordering.
And distinguish the plate from the bed, because they need different purchases. Rotate the sheet 180° and reprint. If the defect stays in the same place on the machine, it is the bed or the probe. If it follows the sheet, the sheet is damaged, and a replacement spring steel sheet is the answer — a dent is below the mesh's resolution and will never be levelled out.
Make Sure It Is Even a Levelling Problem
A perfectly trammed, finely meshed, correctly offset machine still fails on a greasy plate. Fingerprints, PLA residue and release agent all beat geometry. Wash the sheet with soap and water or isopropyl alcohol before concluding anything about the mesh, and match the surface to the material — smooth PEI grips PETG hard enough to take chunks out of itself, which is a surface mismatch rather than a height error. The full diagnostic for that family is why 3D prints won't stick to the bed, and the opposite failure is in why 3D prints get stuck on the build plate.
Read the First Layer, Then Act
| What you see |
Where it comes from |
| Uniformly thin or translucent everywhere |
Z-offset too high |
| Uniformly glassy with squeeze-out |
Z-offset too low |
| One corner wrong, consistently |
Tram, mesh density, or extrapolated edge point |
| Good at the edges, poor in the middle |
Plate crown or dish the mesh is interpolating across |
| One fixed spot wrong, follows the sheet |
Damaged sheet — replace it |
| Fine for ten minutes, then drifts |
Mesh probed cold; soak and re-probe |
| Different every print, no change made |
Probe repeatability |
| Lays down well but lifts later |
Adhesion, temperature, draughts — not levelling |
Work Through It in This Order
- Clean the nozzle hot and wash the plate. Both before anything is measured.
- Run a probe repeatability test. Ten readings on one point. Fix a spread over 0.02mm mechanically before touching the mesh.
- Heat soak, then probe. Printing temperatures, ten minutes, every time.
- Tram by hand, using the mesh as the guide for which corner to move.
- Raise the mesh density and re-probe.
- Set the Z-offset last, with a feeler gauge, then verify it with a measured single-layer patch.
- Print one full-plate single-layer square and photograph it. That image is your baseline, and it makes the next change assessable instead of remembered.
When the Machine Really Is the Limit
Some beds are beyond it. A plate warped past the compensation range, a frame that moves under its own gantry, or a probe design that cannot be made repeatable are all cases where the parts list starts to approach the price of the machine. The relevant specifications are published per model and they are not the headline ones: the probe type, how many points the mesh samples, whether the bed can be trammed mechanically at all, and the flatness tolerance quoted for the plate.
QIDI's enclosed machines and ELEGOO's US store both publish those numbers per model, which makes them comparable in a way that "fully automatic levelling" never is — read them per model rather than per brand. Machines that probe through the nozzle remove the probe-offset problem rather than merely automating around it, and that is the specification worth paying for if levelling is what has been costing you evenings. The slower drift that makes a once-good machine print badly is covered in why 3D printers get worse over time.
When It Is Not Worth Another Evening
There is a version of this where the answer is not a purchase. You need a handful of parts, flat and dimensionally right, and the hours you would spend chasing a 0.05mm mesh error cost more than the parts do. Shops are listed by process and location in the 3D printing directory, and two or three quotes tell you quickly whether your first layer was ever the cheapest thing to fix.
For everything else: the probe measures shape, you set height, and the surface is a third problem entirely. Keep them separate and the first layer stops being mysterious.
Hero photograph by Kadir Celep on Unsplash.
