Weak Across Layers: How Much, and What Moves It

Printed parts are weaker across layers than along them. Everyone agrees on that. The number is another matter: the folklore says an upright print has half the strength, the studies say anything from a tenth to nine tenths, and every one of them is a real measurement. The direction is physics. The number belongs to your temperature, your fan, your material and your test, and there is a coupon for it.

When a bead lands on the layer below, two polymer surfaces meet. One is at extrusion temperature. The other has been cooling for however long the previous layer took. The bond between them is not glue and not a weld in the metal sense; it is molecular diffusion. Polymer chains from each side wander across the interface, and the joint gains strength with how hot the interface is and how long it stays that way. The first serious study of this treated it as a heat problem, measured the temperature of deposited filaments, and found that the fabrication strategy and the envelope temperature around the part had the strongest effect on how well the filaments bonded.^[1] On an open printer the interface is above the glass transition for a few seconds, and the bond is whatever diffusion managed in that time.

That is the first reason a Z-loaded part is weaker: the joint is a partial weld. There is a second reason that is easy to miss. Adjacent beads are rounded, so where two layers meet there is a valley running around the entire part at every layer line. Under tension across the layers, that valley is a notch, and cracks start in notches. A part loaded across its layers is therefore a stack of incomplete welds with a built-in stress concentration at every one.

The two effects add, and they explain the shape of the results. The same material printed the same way is far weaker in tension across layers than in compression, because a notch does nothing under compression: the original ABS study found compressive strength at 80 to 90 percent of the injection-moulded figure while tensile strength varied wildly with how the beads were laid.^[2] Loaded along the beads, the beads carry the load end to end and the bond only sees shear. Loaded across them, the bond carries everything.

Table: Measured strength across bead or layer bonds, as a share of strength along the beads. Each row is a different material, machine and test.

The spread from a tenth to two thirds is not measurement noise. Each row is right about its own conditions. The ratio is a property of bond quality; bond quality is a property of the interface's temperature history; and the temperature history is set by the printer, the profile and the material, not by the part. One machine at one temperature gave a 2:1 change in the ratio across a sweep of nozzle settings.^[6] A rule of thumb cannot survive that.

What every row agrees on is the direction and the rough order. Across-layer strength is always lower, usually by a third or more, and it is the number most likely to move when something about the print changes.

Everything that changes how far a bond gets acts on the temperature of the interface, how long it stays there, or how much material is in contact.

Nozzle temperature. The biggest lever, and not a monotonic one. In the sweep above, PLA's across-layer strength doubled between 190 and 230 °C and then fell again above 250, with the author suspecting moisture and degradation at the top end; PETG did the same with its peak at 245.^[6] The Prusa knowledge base's first fix for layer splitting is the same: raise the temperature in 5 °C steps until the layers hold.^[7] Run bond-critical parts at the hot end of the material's range, not past it.

Cooling. The part fan that saves overhangs and bridges is the thing that quenches the interface. That is a real conflict: the bridging article in this series wants the fan at full, and this one wants it off. Prusa's advice for ABS is to turn the fan off entirely and print in an enclosure, because cold air on a warm part is how layers split.^[7] For PLA, less fan on the layers that matter is a slicer setting, and it costs overhang quality to buy bond quality.

Envelope temperature. A warm chamber keeps every interface above the glass transition for longer, which is why the original bonding study found it to be the dominant variable,^[1] and why materials that are printed in a chamber bond well and bridge badly. Same physics, opposite wishes.

Layer height. The literature points both ways, and honestly so. For PLA printed upright, thicker layers improved the mechanical properties, and layer thickness mattered far more upright than flat.^[3] In another study on in-plane specimens, thinner layers were stronger.^[5] Both can be true: a thicker layer delivers more heat to the interface below it and there are fewer interfaces per millimetre, while a thinner layer packs the beads more closely and leaves smaller voids. Next to temperature, layer height is a second-order lever whose sign depends on what else moved.

Speed and flow. Slower deposition and slightly more material give the beads a wider, more uniform contact and fewer voids at the interface; one study found feed rate and nozzle temperature to be the two parameters that most affected interlayer bonding, with 40 mm/s and 220 °C the best of its range for PLA.^[5] Raising the extrusion multiplier a few percent is the other standard fix for splitting.^[7]

Material. PLA welds to itself well. PETG at its best reached about 60 percent of its flat strength in the sweep, and did not improve past that with temperature.^[6] ABS and ASA need the chamber. The rest of the materials are less measured than the folklore about them suggests.

The layer direction is chosen in the slicer, after the CAD file is closed, and no dimension in the model records it. It is nonetheless the highest-leverage structural decision available once the geometry exists, because it decides which features carry their load along the beads and which carry it across a stack of partial welds.

Table: Features that put load across layers, and the move that turns it.

The first row is the one that costs people parts. A snap-fit or living hinge is a bending feature, and bending across layers is the worst load a printed part can see. If nothing else in this article sticks, that should: anything that flexes gets printed so the flex is in the plane of the bed.

Because the decision lives in the slicer, it also needs to live in the production record. A part that has to be printed on its side is a part whose STL is not enough to reproduce it.

This is the best-measured rule in the series, and the measurements still do not give a number.

There is no standard test for a printed part's across-layer strength. Most studies print a tensile dogbone in two orientations and pull it, which is reasonable, but a printed dogbone's layer lines are notches on its own surface, so every upright result mixes the strength of the bond with the sensitivity of the material to the notch. The two are never separated, and the ratio moves with the dogbone's size, the layer height, the raster pattern inside it, and whether the outer walls were printed with the fan on. That is why one careful sweep can span a factor of two while changing a single setting.^[6]

The effect of layer height on across-layer strength has been reported in both directions.^[3][5] Fatigue and impact across layers are less measured still, and those are the loads that actually break clips and hinges. And nearly everything published is PLA or ABS; the materials people reach for when strength matters are the ones with the least data.

What this means practically: the direction is certain, the mechanism is understood, and the levers are known. The ratio is yours to measure. The coupon does not report a stress; it reports whether a bar of your material, at your settings, snaps across a layer line like a biscuit or bends first, and that is the fact a design needs.

The cooling conflict runs through this whole series. How Far Can a 3D Printer Bridge? and The 45° Overhang Rule, and When It Is Wrong both want the fan at full; this article wants it off. The other two design rules, How Thick Should 3D-Printed Walls Be? and Clearances for 3D-Printed Parts That Have to Move, end the way this one does, in a number that belongs to the machine.

For snap-fits and the orientation they need, see Threads and Snap Fits That Print. For rounding the root of a post or the inside of a corner when there is no fillet tool, see Fillets and Chamfers Without a Fillet Tool. For recording the orientation and profile a part was made with, so that it can be made again, see Versioned Production: Presets, Reorders, and Change Control.