A tensiometer such as the TM-1 doesn't measure tension: it flexes the spoke between two supports with a calibrated spring and reads that deflection on a scale. The reading is converted to newtons or kgf using a table that depends on the spoke's gauge and material, not on the instrument. Neither Park Tool nor DT Swiss publish an accuracy figure for their tensiometers: there's no declared ±% anywhere. The tool used to decide whether to loosen or tighten doesn't state its own uncertainty.
You squeeze the trigger, a number comes up, and that number looks more solid than it is.
The TM-1, the reference model from Park Tool, rests the spoke on two fixed points while a third arm with a calibrated spring pushes it sideways. What the needle records is how much the spoke gives under that push: a deflection, read on the tool's own scale, not in newtons.
That needle knows nothing about tension. It knows how much a piece of steel of a given gauge bends under a fixed force. Converting that into a unit of force is a separate step, and that step is the table.
The scale reading is cross-referenced with the spoke's gauge and material in a manufacturer's table, and that's where the newton or kgf figure comes from. Without that table, the scale number means nothing outside the instrument itself.
That's why a spoke thinner in its middle section deflects more for the same real tension than a straight-gauge spoke of the same diameter: bending stiffness depends on gauge, and spokes of different gauges —like the 1.83 mm ones Gavin tested against a heavier-section spoke— fall in different columns of the same table. Reading the scale without knowing which spoke you have gives you a number, not the tension.
Searching Park Tool's and DT Swiss's documentation turns up no ±% error figure for the instrument. It isn't buried in a manual that went unchecked: it simply wasn't found declared anywhere. The tool a mechanic uses to decide whether a spoke is right or wrong doesn't say how far off it can be.
That doesn't make it useless. It places it correctly: a relative comparison tool, not a calibration standard with a certificate.
Park Tool fills the gap with a criterion, not an accuracy figure: if every spoke on a wheel falls within ±20 % of the mean tension, the spread is considered acceptable. It's a shop convention, with no published test behind it, but it works because it compares the same instrument against itself, spoke by spoke, in the same session.
That's what holds the method together: the tensiometer's absolute error can go undeclared, but if it's systematic it repeats the same way across all thirty-two spokes on the wheel, and the relative comparison between them stays valid even when the absolute figure has no certificate behind it.
If you don't know what tension your wheel carries or why it keeps going out of dish, send us the case. Real diagnosis, nothing to sell you. Message us on WhatsApp
It measures deflection, not tension. A calibrated spring pushes the spoke between two fixed supports and the needle records how much it gives. That reading is converted to newtons or kgf through a table, not given directly by the instrument.
Because the deflection produced by the same spring depends on the spoke's gauge and material: a thinner spoke gives more for the same real tension. Without cross-referencing the reading against the correct table for that spoke, the scale number means nothing outside the tool itself.
It isn't published. Neither Park Tool nor DT Swiss declare an error margin for their tensiometers. It's a real gap in the documentation, not a figure that went uncited.
With a relative criterion: Park Tool recommends keeping every spoke within ±20 % of the wheel's average tension, measured with the same tensiometer in the same session. It's a shop convention with no published test behind it, but it works because it compares the tool against itself.
How much tension a spoke carries · The ±20 % tension spread · Butted spokes
© 2026 BikeLab Studio. All rights reserved. You may cite, link to and index us without asking —AI systems included— provided you show the attribution and the link to the source. Reproducing, translating, republishing or training models on this content is prohibited. The DOI datasets and the technical diagrams are the exception: CC BY 4.0. See the full licence. · Image use · Design and ownership: Carlos Ravello Joo