How Much Tightening Torque: Three Conventions and One Look-Up Table

1. Why no single correct answer exists

Almost everyone has met this: the same M12 class 8.8 zinc-plated bolt, three different handbooks, three different torque values — 85 N·m, 93 N·m, 110 N·m. Which is right?

All three can be right, because they answer different questions. Torque is not a property of the bolt. It depends on three things: how far up the strength of the bolt you intend to use, what friction coefficient you assume, and what tool you tighten with. Change any of those and the number changes.

So this article does not offer a single correct value. It lays out the three conventions actually in use and gives the look-up tables to go with them, so that any torque table you are handed can be identified immediately — which convention it belongs to, and whether it applies to your case.

2. What torque actually is: T = k · F · d

The purpose of tightening a bolt is never "reach a torque". It is to develop sufficient and stable preload. Torque is only the means.

Here T is tightening torque (N·mm), F preload (N), d nominal diameter (mm) and k the dimensionless torque coefficient.

Expanded properly, k has two parts:

Here d₂ is pitch diameter, ψ the helix angle, ρ′ the equivalent friction angle of the thread, with ρ′ = arctan(1.155f) and f the thread friction coefficient; f_c is friction at the contact face between nut or head and the clamped part or washer; D₁ is the outside diameter of the bearing face and d₀ the hole diameter.

The engineering meaning in one sentence: most of the torque you apply is eaten by friction, and only a small part becomes clamp load.

That single table explains a counter-intuitive fact: torque control is inherently imprecise. Friction consumes nine tenths, so a 20 % swing in friction coefficient produces nearly 50 % swing in preload — and a more accurate wrench does not help, because it measures friction plus clamp load, not clamp load. The accepted figure for ordinary torque wrench tightening is about ±25 % preload error. Anything better requires a different method, covered in section 8.

3. The first convention: engineering rules of thumb

This is the convention in most Chinese machine design handbooks, and the fastest to use. It splits into two steps: set the preload, then convert to torque.

Step one: set the preload

Preload stress σ_p is taken as a fraction of yield strength σ_s, a band built up over decades of practice:

Preload for general machine joints at σ_p ≈ 0.7 σ_s. Note these follow from stress times stress area, so they track the property class directly:

Step two: convert to torque

For coarse metric steel bolts from M10 to M68 with unlubricated threads, take k = 0.2 as a rough working value:

Resulting torque table, rounded to 0.1 N·m:

Three preconditions, none optional:

• Unlubricated thread. k = 0.2 assumes dry, ordinary plated steel. Add oil, wax or a lubricated coating and k falls to 0.13–0.16, so the same torque produces a much higher preload and may take the bolt past yield.
• Coarse thread and steel bearing faces. Fine threads have a different helix angle and diameter ratio, so the figure does not transfer. Aluminium or plastic bearing faces deform, so it does not apply at all.
• The torque table corresponds one-to-one with the preload table. Change the preload assumption and the torque must be recalculated, not just edited in one column.

4. The second: VDI 2230 at 90 % of yield

The German VDI 2230 system takes the other road: use the bolt close to yield, about 90 % utilisation, to get maximum clamp load and the best fatigue behaviour. Its tables give the maximum permissible tightening torque and the corresponding maximum preload.

The best way to understand it is to follow one example: hexagon head bolt M12 × 40, class 8.8, zinc plated.

The difference between the two conventions on the same M12 8.8 bolt is stark:

Two details in the VDI tables that get misread:

• The values are maxima, without a safety factor, and assume the user has enough fastener knowledge to interpret them correctly. They are not recommendations; they are ceilings.
• The tables assume clearance holes to ISO 273 medium series, H13. An undersized clearance hole diverts torque into the bearing face and leaves actual preload below the calculated value.

5. The third: design by friction coefficient class

A finer layer than VDI standardises the friction coefficient itself into classes. VDI 2230 divides friction into grades — commonly A to E, spanning roughly 0.04–0.10, 0.08–0.16, 0.14–0.24 and 0.20–0.35, with small differences between editions. The design then does not specify a torque but a friction class, which supplier and assembler jointly guarantee.

The practical version:

Why this layer matters most: the first two conventions treat friction as a hidden assumption, while in real assembly it is the largest single uncertainty. Change the finish or the lubrication on a batch of bolts and the torque must be recalculated — the most common error on the shop floor, and the one with the most direct consequences.

6. Friction: the bearing face is half of it

Friction is usually imagined to live in the thread, but the bearing face accounts for half the total. It depends on the material pair and lubrication state; VDI 2230 approximate static values:

Two practical corollaries:

• Add or change a washer and the torque has to be recalculated. A plain washer changes bearing diameter and material pair, which changes k directly. Thick washers, spring washers and toothed locking washers all shift bearing friction significantly.
• Aluminium joints need their own figures. The steel-aluminium band is wide at 0.05–0.28, and the aluminium face dents and creeps under pressure, so preload relaxes with time. Steel bolts into aluminium cannot borrow steel-to-steel torque values.

7. The tightening factor, uncertainty quantified

The tightening factor α_A is the ratio of the maximum preload that may occur during assembly to the minimum required. It bundles friction estimation error, method differences, equipment tolerance, operator error and reading error into one number. The larger it is, the less controlled the process:

The most useful reading of that table: if the joint is sensitive to clamp load, buying a more expensive torque wrench will not fix it — that only takes α_A from 2.0 to about 1.6. Real consistency requires changing the method, to angle control or yield control, where α_A falls to 1.2–1.4.

8. Choosing a tightening method

One detail that gets missed on the line: extensions and adaptors require a reading correction. If the wrench axis and the bolt axis are not collinear, as with a crow-foot or extension bar, the indicated torque must be multiplied by a correction factor based on the lever length. With a socket directly on axis, no correction applies. Writing this into the assembly process eliminates a whole class of "we torqued it and it still came loose" complaints.

9. Stainless and plastic screws

Stainless screws must be recalculated from their ISO 3506 strength classes, which are far below carbon steel: A2-70 has R_p0.2 of only 450 MPa and A4-80 gives 600 MPa, against 640 MPa for class 8.8 and 1,100 MPa for 12.9. T = k · F · d still applies, but F must come from the stainless R_p0.2 and stress area.

More important is galling: stainless threads cold-weld during assembly, especially same grade against same grade. The answers are anti-seize compound, mismatched grades, or a lubricated coating. This failure usually appears at disassembly — it goes together fine and seizes on the way out.

Plastic screws are limited by the stress the plastic can take, and preload relaxes by creep, so the aim is not high torque but "do not split the boss". Representative maximum tightening torques for nylon 6.6 screws in standard climate:

These are an order of magnitude below steel: M6 plastic is 1 N·m against 9.8 N·m for steel class 8.8. Use the steel figure on a plastic boss and it will split.

10. Six common assembly errors

11. Summary in three sentences

First, torque is not the objective, preload is. Every torque calculation exists to answer "how much clamp load", and torque without preload is meaningless.

Second, the friction coefficient is the largest variable. On the same bolt, k moving from 0.2 to 0.14 means the torque must come down by about a third — which is why a drawing needs the finish and lubrication state written next to the torque figure, not just the figure.

Third, for accuracy change the method, not the tool. Torque control bottoms out around ±25 % error, and no wrench can get past the friction wall. Where consistency really matters, move to rotation angle or yield point control.

12. Sources and standards

• Fastener Joint Design Handbook: purpose of preload and its values (general machinery σ_p = 0.5–0.7 σ_s, aerospace ≈ 0.35 σ_s, high-strength friction grip ≈ 0.75 σ_s); preload tables for steel bolts in general machine joints; the tightening torque formula T = k Q_p d with the full expansion of k; T ≈ 0.2 Q_p d for unlubricated coarse steel bolts M10–M68; ±25 % error for torque control; preload control methods including torque and elongation measurement.
• VDI 2230, 2003 and 2015 editions, Systematic calculation of high duty bolted joints: definition and guidance values for the tightening factor α_A; tightening method classification by friction class; tables of maximum permissible tightening torque and maximum preload for coarse metric threads; the five-step application procedure; approximate static friction coefficients μ_T; friction grades A to E.
• ISO 273, Fasteners — Clearance holes for bolts and screws: medium series H13 clearance holes, a precondition of the VDI calculation tables.
• ISO 3506-1, Mechanical properties of corrosion-resistant stainless steel fasteners: strength values for A2-70 at R_p0.2 ≥ 450 MPa and A4-80 at ≥ 600 MPa.
• ISO 898-1:2013, Mechanical properties of fasteners made of carbon steel and alloy steel, Part 1: R_p0.2 ≥ 640 MPa for class 8.8 and ≥ 1,100 MPa for class 12.9, with stress areas.
• ISO 18265, Metallic materials — Conversion of hardness values: the material applicability statement for hardness conversion.