Your prints have diagonal ghosting that alignment and input shaping couldn’t fix because one belt is 25% looser than the other. In a CoreXY machine, mismatched belt tension behaves exactly like a mechanical skew — the toolhead moves in a slight parallelogram instead of a square, and no amount of software compensation can clean it up completely. Belt tension isn’t just about preventing skipped steps. It’s about matching the kinematic pairs so the machine moves the way the firmware thinks it moves. Here’s how to get it right.
Why Belt Tension Matters More Than Most People Think
GT2 belts are fiberglass-reinforced polyurethane with a steel core. At the correct tension, the steel cords carry the load elastically — they stretch slightly under force and snap back. At too-low tension, the belt rides up on the pulley teeth during direction changes, producing backlash that shows up as surface ringing opposite to the input shaper’s correction. At too-high tension, the steel cords permanently yield, the belt stretches unevenly over its length, and the increased radial load on motor bearings accelerates wear that eventually produces Z-banding.
The sweet spot for a 6mm GT2 belt on a typical 3D printer is a fundamental frequency of 90-110 Hz when plucked like a guitar string at the midpoint of its longest unsupported span. Above 120 Hz, you’re over-tensioning. Below 70 Hz, you’re leaving backlash on the table.
Frequency-Based Tuning: The Reliable Method
You can measure belt tension by frequency without buying anything. Download a guitar tuner app on your phone (GuitarTuna, gStrings, or any chromatic tuner). Pluck the belt at the midpoint of its longest free span between the motor pulley and the idler. The app reads the fundamental frequency. Adjust the tensioner until both belts in a CoreXY pair read within 3 Hz of each other.
For Cartesian machines (Ender 3, Prusa-style), you only have one belt per axis, but the same frequency target applies. The X-axis belt on an Ender 3 should pluck at 90-100 Hz across the span between the motor pulley and the right-side idler.
The numbers per printer type:
| Printer Type | Belt Span (approx) | Target Frequency | Too Loose | Too Tight |
|---|---|---|---|---|
| CoreXY (Voron 2.4, VzBot) | 300-350mm | 95-105 Hz | <80 Hz | >120 Hz |
| Cartesian X-axis (Ender 3) | 300mm | 90-100 Hz | <75 Hz | >115 Hz |
| Cartesian Y-axis (Ender 3) | 350mm | 85-95 Hz | <70 Hz | >110 Hz |
| CoreXZ (Voron Switchwire) | 400mm+ | 80-90 Hz | <65 Hz | >105 Hz |
Longer belt spans have a lower natural frequency at the same tension — a 400mm span at 90 Hz is tighter than a 300mm span at 90 Hz. The frequency-to-tension relationship follows f = (1/2L) × √(T/μ), where L is span length, T is tension, and μ is the belt’s linear mass density. For a typical GT2 6mm belt with μ ≈ 0.0083 kg/m, 90 Hz at 300mm span corresponds to roughly 24N of tension.
CoreXY Belt Matching: The Parallelism Problem
CoreXY kinematics use two belts that must be mechanically identical in tension for the toolhead to move in straight lines. If belt A is tighter than belt B, the toolhead parallelogram distorts — a commanded pure X movement produces a slight Y component, and vice versa.
The symptom: diagonal lines in your prints that should be perfectly vertical or horizontal. Single-perimeter calibration cubes show it — measure the diagonals with calipers. If the two diagonals differ by more than 0.1mm on a 30mm cube, your belts are mismatched.
The fix: tension both belts to the same frequency within 3 Hz. Pluck each belt at the same span location. Adjust the tensioner screws in small increments (quarter-turn at a time) and re-check. This is tedious — it typically takes 5-10 iterations per belt to converge on a match. The result, however, eliminates the single largest source of CoreXY geometric error.
Belt Tensioning Hardware: What Works
Spring-loaded tensioners (Voron-style) apply constant force regardless of belt stretch. The spring maintains tension as the belt beds in during the first 50 hours of printing. The downside: a spring that’s too stiff over-tensions, and one that’s too soft doesn’t maintain tension under acceleration. The stock Voron tensioner spring at 40% compression applies roughly 25N — right in the sweet spot — but aftermarket springs vary wildly.
Screw-adjusted tensioners (Ender 3 style) are simpler and cheaper but require periodic re-adjustment as the belt stretches. Check tension every 200-300 print hours. The Ender 3’s T-nut tensioner on the X-axis is notorious for slipping — the T-nut gradually backs out from vibration. Threadlock (blue, not red) on the tensioner screw eliminates this.
Manual pull-and-clamp (Prusa MK3/MK4 style) relies on the assembler pulling the belt to tension and clamping it in place. This is surprisingly consistent if you use a belt tension gauge, but without one, human feel varies by 30% or more between builds. The Prusa-specific printed tension gauge tool is worth printing — it measures belt deflection under a known spring force and is more consistent than the pluck method for repeated adjustments.
Signs Your Belt Tension is Wrong
Too loose:
– Layer shifting on direction changes (belt skips teeth on the pulley during fast direction reversals)
– Surface ringing at low frequencies that input shaper doesn’t eliminate (belt resonance, not frame resonance)
– First layer dimensional inaccuracy in one axis (X and Y move different distances for the same commanded steps)
– Belt visibly sagging between pulley and idler at rest
Too tight:
– Motor overheating (stepper current increases to overcome belt friction on the pulley)
– Bearing noise — a high-pitched whine during moves that gets worse over time
– Horizontal banding at regular intervals (uneven belt stretch creates a cyclic error as the stretched section passes over the pulley)
– Belt teeth wear prematurely — the reinforcing fibers start showing through the polyurethane at the tooth root
Mismatched (CoreXY):
– Diagonal lines are not perpendicular — a calibration cube’s diagonals differ by more than 0.2mm
– Circles print as slight ellipses
– Surface finish differs between diagonal moves — one direction is clean, the other has ringing
Common Mistakes & How to Avoid Them
Mistake 1: Tightening belts until they “feel right.”
The consequence: humans over-estimate belt tension by 40-60% on average. Most first-time builders over-tension to 140-160 Hz without realizing it. The fix: use a frequency measurement. A guitar tuner app is free and more accurate than any human. If you’re building multiple machines, a dedicated belt tension gauge (Gates 505C or the printed Prusa tool) pays for itself in bearing replacements avoided.
Mistake 2: Not matching CoreXY belt lengths.
The consequence: the two belts stretch differently over time even at the same frequency, because one belt has a longer total path. The fix: always cut CoreXY belts from the same continuous length of GT2 stock. Don’t use belts from different batches — manufacturing tolerances on the steel cord tension vary by 5-10% between production runs.
Mistake 3: Overtightening to eliminate surface artifacts that are actually caused by something else.
The consequence: you chase belt tension when the real problem is loose V-rollers, a wobbly hotend mount, or an extruder with inconsistent tension. The fix: before adjusting belt tension, verify that the toolhead has zero play when grabbed and wiggled, and that the hotend mount doesn’t flex under moderate finger pressure. Rule out mechanical play before touching belt tension.
Mistake 4: Forgetting to re-tension after the break-in period.
The consequence: new belts stretch 2-5% in the first 50-100 hours of printing as the steel cords settle into the polyurethane matrix. Your perfectly tuned 95 Hz belts drift down to 70 Hz over a month of printing, and backlash creeps back in. The fix: check belt frequency after the first 50 hours of printing and re-adjust. Check again at 200 hours and every 300 hours thereafter.
Mistake 5: Using a belt with damaged teeth.
The consequence: a belt that skipped once and stripped a tooth will skip again at the same point in the rotation every time because the missing tooth creates a gap. The fix: if a belt has ever skipped, replace it. Don’t try to tension a damaged belt — the skip was the symptom, not the root cause, and the damage is permanent.
⚠️ Safety Notice: Proper belt tension reduces the risk of mechanical failure during high-speed printing. A belt that snaps during a print can whip and cause injury or damage to wiring. Always power off the printer before adjusting belts. Verify that tensioner hardware is properly secured — a tensioner screw that backs out during a print can cause sudden belt release.
Layer shifting from loose belts often gets confused with stepper driver issues — our layer shifting diagnosis guide covers the full differential diagnosis between mechanical and electrical causes. For machines already running input shaping, our input shaper calibration guide shows how belt tension affects the resonance graph and what a correctly tensioned machine’s ADXL output should look like.
Genuine Gates GT2 6mm belt stock in continuous lengths is available in our printer parts section — cut-to-length with pre-crimped brass ferrules that won’t slip under tension. The steel-reinforced cords maintain frequency within 5 Hz across 500+ print hours, making re-tensioning intervals dramatically longer than generic Amazon belt stock.