ISO 2768 General Tolerance Calculator

Every general tolerance ISO 2768 gives for the letters in your title block — linear, angular, broken edges, straightness, flatness, perpendicularity, symmetry and run-out — each from the length the standard actually keys it on, with the IT grade the note amounts to and the fact that ISO 2768-2 has been withdrawn.

Last updated 24 Sep 2026

ISO 2768 general tolerances

Class letters and lengths → every general tolerance

Linear tolerance class (ISO 2768-1)The small letter in a note such as ISO 2768-mK. It governs every length, width, diameter, centre distance, radius and chamfer on the drawing that has no tolerance of its own.

Geometrical tolerance class (ISO 2768-2)The capital letter. It governs straightness, flatness, perpendicularity, symmetry and run-out where the drawing does not say. Note that this half of the standard has been WITHDRAWN and replaced by ISO 22081 — see the note below.

Nominal size of the dimension (mm)The dimension you want the general tolerance for. ISO 2768-1 steps it from 0.5 mm to 4000 mm; class f stops at 2000 mm and class v does not cover the 0.5 to 3 mm step.

Length of the surface whose flatness or straightness you want (mm)THE PART EVERYONE MISSES. ISO 2768-2’s straightness and flatness are keyed on the feature’s OWN length — the longer side of the surface — and not on any dimension in the linear table.

Shorter of the two sides, for an angle or a squareness (mm)The angular tolerance is keyed on the shorter leg of the angle, and the perpendicularity tolerance on the shorter of the two sides, which is taken as the datum. Both are easy to read off the wrong side of the part.

Longer side, for the linear error an angle allows (mm)An angular tolerance is an angle, so the position error it permits grows with the length of the side it swings. This is where a general tolerance note quietly becomes the loosest thing on the drawing.

Broken edge, chamfer height or external radius (mm)Broken edges have their own, much coarser ladder in ISO 2768-1 — three steps instead of eight, and at 1 mm the tolerance is twice the linear one at the same size.

Not a circuit: the two halves of ISO 2768 as staircases, on two logarithmic length axes that are NOT the same length. The upper panel is the linear tolerance of classes f, m, c and v against the nominal size of the dimension; the lower panel is the straightness and flatness of classes H, K and L against the length of the FEATURE being checked, which has nothing to do with any dimension in the upper table. Each panel has its own pointer, and separating them is the whole point of the figure: reading the geometrical tolerance off the linear length is the commonest way this standard is misapplied. Both are staircases rather than curves because the standard quantises length into steps, and the upper panel's f line simply ENDS at 2000 mm, because class f does not cover the step beyond it.

0.300±mmExample

A drawing marked ISO 2768-mK: a 50 mm dimension, a 100 mm long face, a 30 mm short side and a 1 mm chamfer

Four tables keyed on three different lengths

linear tolerance = f(nominal size, class letter) · angular tolerance = f(SHORTER side, class letter) · straightness and flatness = f(FEATURE’S OWN length, class letter) · perpendicularity = f(SHORTER side, class letter) · run-out = f(class letter alone)

nominal size: the dimension itself, for the linear table. Stepped from 0.5 mm to 4000 mm in eight steps
shorter side: for an angle, the shorter leg; for perpendicularity, the shorter of the two surfaces, which is the datum. Getting this the wrong way round is the most common error in reading the standard
feature’s own length: for straightness and flatness, the length of the line or the longer side of the surface being checked. It has nothing to do with any dimension in the linear table
f, m, c, v: the linear classes, fine to very coarse. f and m share the same angular ladder; c and v share the same broken-edge ladder
H, K, L: the geometrical classes. K is exactly twice H and L exactly twice K in flatness, at every step but the finest

Worked example

A drawing marked ISO 2768-mK: a 50 mm dimension, a 100 mm long face, a 30 mm short side and a 1 mm chamfer
The 50 mm dimension falls in the “over 30 to 120” step, so the m class gives ±0.30 mm — a total band of 0.6 mm, or 600 µm
Put that in perspective with an ISO 286 grade. At Ø50, 600 µm sits between IT13 (390) and IT14 (620). IT11 there is only 160 µm, and an H7 is 25. So the general note is about 24 times looser than an H7 at the same size, and still 3.8 times looser than IT11, which is the coarsest grade anybody specifies a fit to. That gap is what a general tolerance is FOR, and why it must never be relied on for a fit
The angular clause, which is where the surprise lives. The 30 mm short side falls in “over 10 to 50”, so class m allows ±30′, half a degree. Over that 30 mm side that is ±0.262 mm of linear error — comparable to the linear tolerance. But swing it over a 400 mm leg and it becomes ±3.49 mm, which is 12 times the linear tolerance on the same drawing
The chamfer has its own, much coarser ladder. A 1 mm broken edge in class m is ±0.2 mm — twenty per cent of the feature, and 2 times the linear tolerance at the same size. Three steps in the whole ladder against eight in the linear one
Now the K half, and note that NOTHING here is keyed on the 50 mm dimension. Flatness comes from the FACE'S own length: 100 mm falls in “over 30 to 100”, so class K gives 0.20 mm of flatness over that face, and the same number for straightness of a 100 mm line
Perpendicularity comes from the SHORTER of the two sides, taken as the datum. With a 30 mm short side that is the “up to 100” step: 0.4 mm. If you read it off the 300 mm side of a long bracket instead you would get 0.6 mm and believe you had 50 per cent more tolerance than you have
Symmetry in class K is 0.6 mm at this size, and run-out is 0.2 mm — the one value in the whole standard that does not depend on size at all, in any class. Note that class H's symmetry does not depend on size either: it is 0.5 mm at every length
One more thing about the K. ISO 2768-2:1989 is withdrawn, and ISO 22081:2021 replaced it. The m is still live. So this drawing's title block is half current, and the replacement does not have classes at all — it asks the drawing to state a general profile tolerance and the datum system it is referred to, which is the decision the letters used to make for you

ISO 2768-1 linear tolerances, and what the medium class is worth as an IT grade

The last column is the one that puts a general tolerance in perspective, and it is computed rather than quoted: it is the finest ISO 286 grade whose band is at least as wide as the m class’s total tolerance at the middle of that size step. It lands on IT13 or IT14 almost everywhere, which is casting and stamping territory. That is the honest reading of a general tolerance note: it is not a tolerance in the sense a fit is, it is a statement that nobody has thought about this dimension, and it is about four grades looser than the coarsest fit anybody specifies. Two gaps in the table are real and worth knowing. Class f has no entry above 2000 mm and class v has none below 3 mm — so a drawing citing ISO 2768-f with a 3 m dimension on it has said nothing at all about that dimension, and the same is true of ISO 2768-v on a 2 mm one. These dimensions come from a published standard’s table, not from a formula. The standard itself is cited below and the printed values are attributed to the catalogue they were taken from; a different publisher may round differently in the last digit.
Nominal size (mm)	f — fine	m — medium	c — coarse	v — very coarse	The m class as an ISO 286 grade
0.5 to 3	±0.05	±0.10	±0.20	—	IT14
over 3 to 6	±0.05	±0.10	±0.30	±0.50	IT14
over 6 to 30	±0.10	±0.20	±0.50	±1.00	IT14
over 30 to 120	±0.15	±0.30	±0.80	±1.50	IT14
over 120 to 400	±0.20	±0.50	±1.20	±2.50	IT14
over 400 to 1000	±0.30	±0.80	±2.00	±4.00	beyond ISO 286
over 1000 to 2000	±0.50	±1.20	±3.00	±6.00	beyond ISO 286
over 2000 to 4000	—	±2.00	±4.00	±8.00	beyond ISO 286

ISO 2768-2 straightness and flatness, and perpendicularity, in millimetres

Two columns of this table are keyed on DIFFERENT lengths and that is the single most-missed thing about ISO 2768-2. Straightness and flatness are keyed on the feature’s own length — the longer side of the surface you are checking — and have nothing to do with any dimension in the linear table. Perpendicularity is keyed on the SHORTER of the two sides, which the standard takes as the datum. So a 300 × 20 mm bracket in class K gets 0.4 mm of flatness over its 300 mm face and 0.4 mm of squareness from its 20 mm leg — read the squareness off the 300 and you would have given yourself 0.6 and be out of tolerance without knowing. Notice also that the three classes are a clean factor of two apart in flatness, which is how a shifted column was caught here: one published source prints the PERPENDICULARITY row’s 0.5 / 1.0 / 2.0 in the last flatness row, and because the factor of two holds for both readings the ratio could not settle it and a third source had to. These dimensions come from a published standard’s table, not from a formula. The standard itself is cited below and the printed values are attributed to the catalogue they were taken from; a different publisher may round differently in the last digit.
Length (mm)	Flatness H	Flatness K	Flatness L	Squareness H	Squareness K	Squareness L
up to 10	0.02	0.05	0.10	0.20	0.40	0.60
over 10 to 30	0.05	0.10	0.20	0.20	0.40	0.60
over 30 to 100	0.10	0.20	0.40	0.20	0.40	0.60
over 100 to 300	0.20	0.40	0.80	0.30	0.60	1.00
over 300 to 1000	0.30	0.60	1.20	0.40	0.80	1.50
over 1000 to 3000	0.40	0.80	1.60	0.50	1.00	2.00

What ISO 2768-mK actually cites, in 2026

This was checked on ISO’s own catalogue during the making of this page rather than taken from any of the guides that explain ISO 2768, none of which mentions it. The practical consequence is small but real: if a customer or an auditor asks whether your title block cites a current standard, the answer for the linear half is yes and for the geometrical half is no. The same designation can mean different dimensions in different standards families — ANSI against ISO, inch against metric, one national standard against another. The family used here is named beside every figure; check which one your part was made to.
The designation	What it governs	Standing
The small letter — f, m, c or v	Permissible deviations for every linear and angular dimension without an individual tolerance, plus the separate and much coarser ladder for broken edges	ISO 2768-1:1989 is still PUBLISHED. ISO’s catalogue entry lists it as published and confirmed, and flags it for revision. A drawing citing it is citing a live standard
The capital letter — H, K or L	General geometrical tolerances: straightness, flatness, perpendicularity, symmetry and circular and axial run-out	ISO 2768-2:1989 is WITHDRAWN. ISO’s catalogue entry says so in one word and names ISO 22081:2021 as what revised it. So half of “ISO 2768-mK” cites a withdrawn document
Its replacement, ISO 22081:2021	A general geometrical specification and a general size specification, each stated on the drawing with its own value	Published. And it works differently: it has no table of classes at all. Instead the drawing must give a general surface-profile tolerance WITH A DATUM SYSTEM, and a general size specification separately. There is no ISO 22081 equivalent of the letters, which is why nothing on this page can convert one into the other
What to do about it	Serve the drawings that exist and specify the new ones properly	Millions of drawings cite ISO 2768 and will for decades, and this page computes them correctly. For a NEW drawing, ISO 22081 is the current answer for the geometrical half, and it asks you for a decision the old letters made for you — which datums the general tolerance is referred to. That question does not have a default, which is exactly why the old standard was withdrawn

Four ways a general tolerance note bites

None of these four is subtle once it is written down, and all four are routine on real drawings. The third one is the most expensive in practice, because a chamfer that is within its general tolerance can still be too big for the counterbore it has to enter — see the counterbore and countersink calculator for that geometry.
The trap	What actually happens
Reading the squareness off the long side	ISO 2768-2 takes the SHORTER of the two sides as the datum. On a 300 × 20 mm bracket in class K that is 0.4 mm and not 0.6, and the difference is 50 per cent of the tolerance
Forgetting that an angular tolerance is an angle	Class m allows ±30′ on a side between 10 and 50 mm. Over a 400 mm long leg that is 3.5 mm of position error at the far end — ten times the linear tolerance on the same drawing. A general tolerance note is almost always the loosest thing on a drawing, and it is almost always the angular clause that makes it so
Assuming the linear class covers the chamfers	It does not. ISO 2768-1 gives broken edges, external radii and chamfer heights their own three-step ladder, and at 1 mm the medium class allows ±0.2 mm on a 1 mm chamfer — twenty per cent of the feature, and twice what the linear table gives at the same size
Assuming the note covers a fit	It never does, and it should not. The m class at Ø50 is a 600 µm band, between IT13 and IT14 — about twenty-four times an H7. Every fitted feature needs its own tolerance, and the ISO 286 fit calculator is where those come from. The general note exists to cover the dimensions where nothing fits anything

Seven tables, three different lengths, and one withdrawn half

A general tolerance note is not one table, it is seven, and they are keyed on different lengths. The linear tolerance comes from the dimension itself. The angular tolerance comes from the SHORTER leg of the angle. Straightness and flatness come from the FEATURE’S OWN length — the line or the longer side of the surface being checked, which has nothing to do with any dimension in the linear table. Perpendicularity comes from the shorter of the two sides, which the standard takes as the datum. Broken edges have their own three-step ladder that is much coarser than the linear one. And run-out does not depend on size at all. Reading the wrong length into the wrong table is the most common way ISO 2768 is misapplied, and it goes wrong in the expensive direction: a 300 × 20 mm bracket’s squareness comes from the 20 mm side, and reading it off the 300 gives you 50 per cent more tolerance than you have.

What the letters mean, and what they are worth. The small letter — f fine, m medium, c coarse, v very coarse — governs the linear and angular half, ISO 2768-1. The capital letter — H, K, L — governs the geometrical half, ISO 2768-2. Written together they give the string people search for, ISO 2768-mK. It is worth knowing how coarse these really are: the m class at Ø50 is a 600 µm total band, which sits between IT13 and IT14 and is about twenty-four times an H7. A general tolerance is a statement that nobody has thought about this dimension, and it should be read that way. Every fitted feature needs its own tolerance, from the ISO 286 fit calculator.

Half of ISO 2768-mK cites a withdrawn standard. ISO’s own catalogue entry for ISO 2768-2:1989 reads “Withdrawn” and names ISO 22081:2021 as the document that revised it. ISO 2768-1:1989 is still published, confirmed, and flagged for revision. So the m is live and the K is not — and none of the guides explaining the designation says so. This page still computes the 2768 numbers, because millions of drawings cite them and will for decades. But for a new drawing the geometrical half now belongs to ISO 22081, which works differently in a way worth understanding: it has no table of classes, and it requires the drawing to state a general geometrical specification with its own tolerance value AND its own datum system. That datum decision is the thing the old letters made silently, and it is why the part was withdrawn rather than merely renumbered.

The angular clause is usually the loosest thing on the drawing. Class m allows ±30′ on a side between 10 and 50 mm, which sounds modest. Half a degree is 0.26 mm of position error over 30 mm and 3.5 mm over 400 mm, because an angular tolerance is an angle and the displacement grows with the arm. Two things follow. Anything that has to line up with anything else at a distance needs its own control, and it should be a geometrical one with a datum rather than an angular dimension. And choosing the fine class buys nothing on angles at all — ISO 2768-1 gives f and m the same angular ladder, a fact that is easy to miss and that makes the fine class a much worse deal than it looks.

Where this page fits with the rest. A general tolerance covers the dimensions nobody has thought about; the IT grade tolerance calculator says what grade you are implicitly asking for and which process holds it; the ISO 286 fit calculator tolerances the features that actually fit something; the tolerance stack-up calculator adds up whatever you end up with; and the true position and MMC bonus calculator handles the positional controls that a general geometrical class cannot express, because ISO 2768-2 has no position column and never did.

Frequently asked questions

What does ISO 2768-mK mean?

Two separate things joined into one string. The lowercase m is the medium class of ISO 2768-1 and sets the permissible deviation for every linear and angular dimension on the drawing that has no tolerance of its own — ±0.3 mm on a 50 mm dimension, ±30′ on a 30 mm short side. The capital K is the medium class of ISO 2768-2 and sets the general geometrical tolerances: 0.2 mm of flatness over a 100 mm face, 0.4 mm of squareness off a 30 mm datum, 0.2 mm of run-out at any size. They are independent choices and a drawing may use any combination.

Has ISO 2768 been withdrawn?

Half of it. ISO’s catalogue entry for ISO 2768-2:1989, the geometrical part, reads “Withdrawn” and names ISO 22081:2021 as the standard that revised it. ISO 2768-1:1989, the linear and angular part, is still published and confirmed, with a revision flagged. So a title block saying ISO 2768-mK is half current. For a new drawing, ISO 22081 is the answer for the geometrical half — and it will ask you for something the letters never did, namely which datums the general tolerance is referred to.

Does the flatness tolerance come from the dimension?

No, and this is the part most people get wrong. ISO 2768-2’s straightness and flatness are keyed on the FEATURE’S OWN length — the length of the line, or the longer side of the surface you are checking — and have nothing to do with any dimension in the linear table. Perpendicularity is keyed on the SHORTER of the two sides, because the standard takes it as the datum and compensates for a short datum by tightening the tolerance. Run-out is keyed on nothing: it is 0.1, 0.2 or 0.5 mm in classes H, K and L at every size.

Does the general tolerance cover a fitted diameter?

Legally yes, practically never. At Ø50 the m class gives a 600 µm total band, which lands between IT13 and IT14 and is roughly twenty-four times an H7. Any feature that fits something — a bearing seat, a dowel hole, a spigot, a bore for a bush — needs its own tolerance, and the general note exists to cover the other ninety per cent of the dimensions. A useful discipline: if you cannot say what the dimension mates with, leave it to the general note; if you can, tolerance it.

Is class f worth choosing over class m?

Less often than it looks, for two reasons. It buys you nothing at all on angles, because ISO 2768-1 gives classes f and m the same angular ladder — identical at every step. And it buys you nothing on broken edges either, because f and m share that ladder too. What it does do is halve the linear tolerance, and that applies to every unconsidered dimension on the drawing at once, including the ones where it adds cost for no function. Tolerancing the few dimensions that matter and leaving the note at medium is usually the better trade.

What tolerance does a 1 mm chamfer get?

±0.2 mm in classes f and m, and ±0.4 mm in classes c and v — from ISO 2768-1’s separate broken-edge table, which has three steps against the linear table’s eight. That is twenty per cent of a 1 mm chamfer, and twice what the linear table would give at the same size. It matters when the chamfer has to enter something: a chamfer at the top of its general tolerance can be too large for the counterbore or the mating recess it was drawn to suit.

Related calculators

References

ISO 2768-1:1989, General tolerances — Part 1: Tolerances for linear and angular dimensions without individual tolerance indications. Cited by number. Status checked during this batch on ISO’s own catalogue entry: still PUBLISHED, confirmed, and flagged for revision. It is Part 2 that has gone.
ISO 2768-2:1989, General tolerances — Part 2: Geometrical tolerances for features without individual tolerance indications. ISO’s own catalogue entry for this document was fetched during this batch and reads Withdrawn, with “Revised by ISO 22081:2021”. So the K in “ISO 2768-mK” cites a withdrawn standard and the m does not — which is not what any of the CNC-shop guides explaining ISO 2768 say.
ISO 22081:2021, Geometrical product specifications (GPS) — Geometrical tolerancing — General geometrical specifications and general size specifications. The document that replaced ISO 2768-2. It does not carry a table of classes at all: it requires the drawing to state a general geometrical specification with its own tolerance value and its own datum system, and a general size specification separately. That is why there is no ISO 22081 equivalent of “mK” and why nothing on this page can compute one for you.
CSL Industrielle Messtechnik. ISO 2768 tolerance tables — general tolerances f, m, c, v and H, K, L. The source for all seven ISO 2768 tables on this page. Checked cell by cell against RivCut’s and rpProto’s published ISO 2768 charts.
RivCut. ISO 2768 tolerance chart: general tolerances (f/m/c/v). The second source, and the one that found the trap. It agrees everywhere except the last row of the straightness and flatness table, where it prints 0.5 / 1.0 / 2.0 for classes H / K / L — which is the PERPENDICULARITY row’s values. Two other sources give 0.4 / 0.8 / 1.6 and are used. The class ratio K = 2H, L = 2K holds for both readings, so the ratio could not settle it and a third source had to.
rpProto. ISO 2768-1 & 2 general tolerances chart (PDF). The third source, fetched for the single disputed row, which it gives as 0.4 / 0.8 / 1.6 in agreement with CSL.
ISO 286-1:2010, Geometrical product specifications (GPS) — ISO code system for tolerances on linear sizes — Part 1: Basis of tolerances, deviations and fits. Cited by number; the standard is copyrighted and its tables are not reproduced here. What this page uses from it is its STRUCTURE, which is not a table: the standard tolerance factor i = 0.45·∛D + 0.001·D evaluated at the geometric mean of each nominal size step, the grade multipliers (IT5 = 7i, IT6 = 10i, IT7 = 16i, IT8 = 25i, IT9 = 40i, IT10 = 64i, IT11 = 100i and so on), the rule that IT(n+5) is ten times IT(n) from IT7 upward, the formulas for each letter’s fundamental deviation, and the Δ correction for hole letters J to ZC. Every one of those is checked here against a published table rather than trusted.