Press Fit Tolerance, Transition Fit and Clearance Fit

Start from the function, not from a table

Two reading rules make the rest of this page usable. The letter in a tolerance class sets where the tolerance zone sits relative to the basic size. The number, the IT grade, sets how wide that zone is. For most classes those are separate choices, and that separation is what makes the system compact. It is not universal. A few classes are defined in terms of the grade itself, and several letters carry a correction that depends on the grade as well. Those positions have to be read from the table rather than reasoned out.

Capital letters are internal features, which the standard calls holes. Lowercase letters are external features, which it calls shafts. A fit is always a pair and it is written hole first, as in H7/g6. The case is not decoration. It tells you which member of the pair you are looking at, and the same letter in the two cases does not describe the same zone. What the case does not tell you is which side of the basic size the zone falls on. That comes from the letter, and for several letters it moves with the grade as well.

Almost everything else follows from two facts. An H hole is never smaller than the basic size. An h shaft is never larger than it. Hold those two and you can reason your way through a great many fits without reading a single value.

What the standard will not do is tell you whether the joint works. It has no opinion on your load, your temperature, your lubricant, your assembly method or how often the thing has to come apart. Choosing the fit is engineering. The designation is only how you write the answer down.

Choosing a fit, step by step

Work down this list in order. The early questions eliminate most of the search space, and the last one is the only place a letter appears.

1. Ask what the joint is for. There are four answers. It allows relative motion. It locates parts while something else carries the load. It transmits load by friction through the fit itself. Or it holds a bought-in precision component.
2. If it holds a bought-in component, stop and open that component's catalogue. Bearings, seals, bushings and linear rail blocks all carry their own fit recommendations, and those override the general system.
3. If it moves, size the clearance from the lubricant film, the differential thermal growth, the form error on both parts, the contamination you expect and any wear you are prepared to live with. Clearance is a designed dimension, not a leftover.
4. If it only locates, decide what actually carries the load. A key, a spline, bolts or a flange means the fit only has to hold concentricity. That is transition territory, or a light interference at most.
5. If the fit is the only load path, size the contact pressure and check the hoop stress in the outer member. You are doing a stress calculation, not picking a letter off a list.
6. Ask how it comes apart and where. Never, in a factory with a press, or in the field at two in the morning with what is in the van. Those are three different fits.
7. Check the cross-cutting effects before you commit. Operating temperature, rotational speed, dissimilar materials, plating and coatings can each undo a correct designation.
8. Only now choose the classes. Specify the hole one IT grade coarser than the shaft unless you have a reason not to, because a hole is harder to hold to size than a shaft for the same manufacturing effort.

If it has to move: what sets the clearance

Clearance is a functional dimension and it should be arrived at deliberately. Add up the lubricant film you need, the differential growth between the two parts at operating temperature, the form error each part is allowed, the dirt you expect to live with and the wear you will accept before the joint is scrap. That sum is your minimum clearance. The class is chosen to cover it, not the other way round.

Some cases hand you the answer through their own physics. A hydrodynamic journal runs on a film whose thickness, power loss and stability all come out of the clearance. Too much clearance and the film collapses at low speed. Too little and you get heat and then seizure. That is a bearing calculation, and it gives you a target clearance that you then find a class to cover.

Leakage through an annular gap is extremely sensitive to the gap. A valve spool or a plunger clearance is set by permitted leakage, and small changes in the gap make large changes in the flow. Anything running hot, dirty or unlubricated wants more clearance than the same joint running clean and cool.

If it only has to locate: transition fits and what they do not promise

A transition fit is one where the hole and shaft tolerance zones overlap. A given pair can come out with a small clearance or a small interference depending on where each part happened to land inside its own zone. That is the definition, and it carries a consequence most explanations skip.

A transition fit does not have an answer. It has a distribution. And that distribution is not an even split between the two outcomes, which is the assumption people quietly make. Where it lands is set by where your two processes actually run inside their zones, and processes do not sit in the middle. A reamer produces holes that drift across the zone as the tool wears. Machinists routinely bias shafts toward the high end and holes toward the low end, because an oversize shaft or an undersize hole can be recut while the opposite is scrap. In a hole-basis pair both of those biases push the same way, toward interference.

So assembly planning has to work for both outcomes. If the process cannot cope with a loose one, or the function depends on grip, a transition fit is the wrong choice no matter how good the average looks. Use it where you want accurate concentric location with very little radial play, and where something else is carrying the load.

If the fit is the load path: interference, and why more is not safer

An interference fit holds by friction. The interference stretches the hub and compresses the shaft, that elastic recovery produces contact pressure at the interface, and the holding force is that pressure multiplied by the contact area multiplied by the coefficient of friction. Everything else follows from that sentence.

Two scaling facts are worth carrying around, and they come from the same place. The friction force is contact pressure times contact area times the coefficient of friction, and the contact area grows with both diameter and engagement length. So axial holding capacity grows with both. Torque capacity grows with the same product and then picks up one more factor of diameter, because that friction force acts at a radius. At the same contact pressure, doubling the diameter roughly doubles the axial capacity and roughly quadruples the torque capacity.

Now the reason more interference is not safer. The contact pressure that grips the shaft also puts the hub bore into tension, and for a hub with a free outer surface the tensile hoop stress at the bore is larger than the contact pressure that produced it, at any wall thickness. Push the interference up and you are raising a stress that is already the highest in the part. Thin-walled hubs are where this bites hardest, because contact pressure is strongly sensitive to wall ratio when the wall is thin and only weakly sensitive when it is thick. A thin hub gives you little grip for a lot of stress.

Surface finish quietly takes some of your interference away. Pressing flattens the asperities on both surfaces, so the effective interference after assembly is less than the interference you measured before it. The loss depends on the finish of both parts and it does not scale with diameter, which means a small interference fit is proportionally far more sensitive to surface finish than a large one. A shrink or expansion fit avoids the problem, because the surfaces are not sliding against each other under pressure during assembly, so it gives more holding capacity than a cold press at the same nominal interference.

Two more things that catch people out. Pressing a thin-walled bushing into a housing closes its bore. In the thin-wall limit the wall moves inward bodily rather than compressing, so the bore diameter closes by very nearly the full diametral interference. Machining that bore to final size before pressing is therefore a mistake, and the usual answer is to finish it after assembly. And two assemblies that both conform to the same designation can need very different press forces, because one pair may sit at opposite ends of their zones from the other. A press force limit set from one sample is not a process control.

If it holds a bearing: the catalogue decides, not the fits system

Go to the bearing maker's catalogue for that part number and use what it says. This is not deference for its own sake. Rolling bearing bores and outside diameters are toleranced to the rolling bearing standard, ISO 492 for radial bearings, rather than to the general fits system, and both are biased below nominal size. So a shaft class that gives a light transition against an ordinary hole gives real interference against a bearing bore. Reasoning from the general system here gives you a tighter fit than you intended.

The decision logic underneath the catalogue is worth knowing, because the tables leave it implicit. The governing question is which ring carries the rotating load. A ring carries a rotating load when the direction of the applied load sweeps around that ring's circumference. That is not the same as the ring itself rotating, and confusing the two is where most of the mistakes come from. The ring carrying the rotating load is the one that needs the interference.

Work it through. On a rotating shaft under a load that is fixed in space, which is the ordinary gearbox or motor case, the inner ring carries the rotating load and gets the interference. On a stationary shaft inside a rotating housing, such as a wheel hub, the outer ring carries it. And on an out-of-balance rotor, where the unbalance dominates the load and its direction turns with the shaft, the answer inverts relative to the ordinary case and the outer ring is the one carrying the rotating load. Real rotors carry space-fixed loads too, so check which one actually governs before inverting anything.

Get it wrong and the ring creeps. Under a rotating load the ring deflects at the loaded point, that point travels around the seat, and the ring rolls very slowly relative to its seat. The symptom is fretting, reddish-brown oxide debris around the seat, and progressive wear that makes the seat oversize and the problem worse. No amount of axial clamping fixes creep.

The other half of the bearing decision is that interference fitting a ring squeezes it, which reduces the bearing's radial internal clearance. Fit selection and internal clearance class have to be decided together. A fit chosen without reference to the clearance class can preload the bearing into itself, which raises running temperature and shortens life. This is the most concrete form of the more-is-safer error.

If it locates a fixture: dowels are a different problem

Dowel fits get treated as a bearing-seat problem almost everywhere, and they are not one. A bearing seat transmits load and never comes apart. A dowel gives repeatable location and usually has to come apart. That difference drives everything.

The convention is interference in one part and a locational fit in the other, so the assembly can be separated and put back exactly where it was. Press both and you cannot take it apart without damage, and you have also made the two hole positions agree perfectly or fight each other.

Two round pins in two round holes overconstrain the assembly. The centre distance on one part has to match the centre distance on the other exactly, which no process gives you. The usual resolution is to relieve the second pin along the line joining the two holes, so it constrains rotation without also trying to constrain centre distance.

Two practical points that cost real money. Blind dowel holes need a vent, because trapped air stops the pin seating fully and trapped fluid can split the part. And holes reamed before heat treatment move in both size and position during hardening, which is why hardened plates get jig ground or wire cut after heat treatment rather than reamed before it.

Reading the designation you have chosen

The IT grade expresses a level of manufacturing precision rather than a physical width. Two features at the same grade and different sizes are about equally hard to make, and they do not have the same permitted variation. The permitted width grows with size, because deflection, thermal growth, measurement uncertainty, tool wear and workholding distortion all grow with size, but it grows less than in direct proportion. Doubling the size does not double the tolerance.

The system is stepped rather than continuous. Sizes are divided into ranges and one width applies across a whole range, so the permitted variation is flat inside a step and jumps at the boundary. A designer who nudges a nominal size across a step boundary changes the tolerance without changing anything else on the drawing. That is worth knowing and it is almost never taught.

The letters run in bands. Against an H hole, the shaft letters up to and including h give clearance, with the designed clearance shrinking at each step as you move up the alphabet. h is the boundary, and it is the tightest shaft that still cannot produce interference. Around js, j, k, m and n you are in transition territory. From p upward you are into interference. Treat those bands as a map rather than a rule, because the point where transition becomes interference moves with both the nominal size and the grades.

That last point generalises into the single most useful correction on this topic. The fit category is a property of the pair of classes, not of the shaft letter on its own. The same shaft letter gives a transition fit against one hole class and something else against another, because changing the grades changes the zone widths and therefore changes whether the zones overlap at all. Anyone who tells you that a given letter is a transition letter, without naming the hole class and both grades, is guessing.

• H7/h6 is a clearance fit, not a transition fit. An H hole is never below the basic size and an h shaft is never above it, so the zones cannot overlap whatever the grades. Its minimum clearance is zero, and a fit whose minimum clearance is zero is still a clearance fit. This is the most common misclassification in published explanations.
• The descriptive names, sliding, close running, locational transition, medium drive, come from handbook and preferred-metric-fits practice rather than from the text of the fits standard itself. They are useful vocabulary. They are not definitions you can cite.
• A hole one grade coarser than the shaft is convention, not law. Equal grades on both members are perfectly legal. They just cost more for the same result.

Temperature, speed and coatings move the fit after you specify it

A size tolerance is a statement about the part at the standard reference temperature of twenty degrees Celsius, which is set by ISO 1, and at no other temperature. Everything in this section is about what happens between that statement and a machine running in a real building.

Two parts of the same material at the same uniform temperature keep the same fit in the way that matters. Every dimension scales by the same factor, the interference scales with them, and so the contact pressure is unchanged. The absolute gap moves a little. The character of the fit does not. Dissimilar materials do not have that protection. A steel shaft in an aluminium housing loses interference as the assembly heats, because aluminium expands faster than steel. That has to be checked at the operating temperature rather than on the bench. Worse, repeated thermal cycling of a dissimilar-material interference fit can loosen it progressively, because the parts slip a little on each cycle and do not fully return.

Speed does the same thing by a different route. A rotating hub grows radially in proportion to the square of its angular speed, and that growth eats the interference you have at rest. A fit that is correct stationary can be marginal at running speed.

Finally, remember that surface treatment happens after machining. Electroplating and hard anodising change the finished dimensions. A coated shaft finishes larger than it was machined and a coated bore finishes smaller. Specify where in the process the dimension applies, or you will get an argument about it later.

Both parts were in spec and they still would not go together

This is the bridge from design into inspection, and it is where two-point measurement quietly lets you down.

The first cause is which rule your drawing runs under. In the ISO scheme the default is the principle of independency, stated in the 2011 revision of ISO 8015, so size and form are independent requirements unless the envelope requirement is specifically invoked. Older drawings may not carry that default, so check what the title block invokes. Under ASME Y14.5, rule one requires perfect form at the maximum material condition for a regular feature of size, with form error permitted to grow as the feature departs from that condition. Same nominal drawing, different meaning. A perfectly conforming ISO part can carry form error that an equivalent ASME part could not.

The second cause is lobing. An odd-lobed bore, a three-lobed one being the classic, can measure a constant diameter with any two-point instrument in every orientation and still refuse to accept a plug of that diameter. The part passes every micrometer check you can devise and fails to assemble. Odd lobing is characteristically produced by centreless grinding and by three-jaw chucking, so it is not exotic.

Verifying the fit: what each kind of check actually proves

There are two kinds of check and they produce different evidence. Attribute checks give you a decision and no number. Variable checks give you a number. Which you use changes what you can detect later, so this is a bigger decision than it looks.

Limit gauges are the attribute case. Under Taylor's principle the GO member checks the maximum material limit over the full length and full form of the feature, so it evaluates size and form together, while the NO-GO member checks the least material limit at individual cross sections with effectively two-point contact. That is why a bore which passes a GO plug is not merely within size, it is within size along its whole length including any bend, taper or lobing. Note that this is the principle rather than a description of every gauge on the shelf. Real NO-GO plugs and rings are often made full form rather than as the two-point check the principle asks for, which means they can be blocked by form error and pass a part that a true two-point check would have caught.

The variable case runs on a capability ladder. At coarse grades handheld instruments are genuinely fine. In the middle band, micrometers, dial bore gauges and limit gauges all do the job, and this is where most shop practice sits. At fine grades everything becomes comparative. A bore gauge or an air gauge does not measure a bore, it measures the difference between that bore and a master, so the whole accuracy of the check is inherited from a setting ring or a stack of gauge blocks. At the finest grades you need a temperature-controlled room, soaked parts and a real uncertainty budget, and without those you cannot verify the grade honestly no matter what instrument you buy.

Air gauging deserves a mention because quality managers often do not know it is an option. It is non-contact, so it does not wear the part, it is fast enough for production, and it works well in tight bores. The trade is that it is comparative over a narrow range around a setting master, it needs a master for every size, and it responds to surface texture as well as to size.

For a CMM, be careful what the specification means. A maximum permissible error figure from an ISO 10360 acceptance test is a machine specification obtained under defined test conditions. It is not the task-specific uncertainty of your measurement. Establishing that requires comparison against calibrated workpieces, which is the method set out in ISO 15530-3.

• GO gauges wear in the direction that increases what they accept. A GO plug gets smaller and starts passing undersize bores. A GO ring gets larger and starts passing oversize shafts. Both wear modes fail toward false acceptance, which is the dangerous direction.
• An attribute gauge produces no measured value, so nothing in your per-part data can show that the gauge has drifted. There is an indirect signal at process level, because wear moves the reject fraction in a predictable direction, and charting that fraction or re-checking rejected parts on a variable instrument can raise suspicion. It is lagging, it is easily confounded with a genuine process change, and it never tells you which parts were affected. The calibration record is the direct evidence.
• Limit gauge tolerance and wear allowance are conventionally set as fractions of the workpiece tolerance. The same physical amount of wear therefore consumes the allowance faster on a gauge used for a fine-grade feature than on one used for a coarse feature.

What measurement uncertainty does to the accept or reject decision

Here is the pivot. As the grade tightens, the tolerance shrinks while the uncertainty of your measurement process stays exactly where it was. So the ratio between them falls purely because someone changed the drawing. At some point the gauge is no longer capable of making the decision the drawing is asking it to make, and nothing about that is visible in the measurement itself.

The number that tracks this is the test uncertainty ratio. It takes the tolerance of the characteristic being judged, as a half width, over the expanded uncertainty of the measurement process as you actually perform it. That denominator is the part people get wrong. It has to include the setting master, the comparator, resolution, repeatability, reproducibility across operators, thermal effects and the workpiece itself. Putting the instrument's datasheet accuracy in there instead gives you a test accuracy ratio, which is a different and more flattering number.

On the conventions, be precise about where they come from. The familiar four to one ratio derives from United States military calibration practice and is carried into ANSI/NCSL Z540.3 as one route to satisfying its false accept risk limit. The older gauging rule of ten came first, and the military ratio was a deliberate relaxation of it on cost grounds. The gaugemaker's tolerance applied when limit gauges are made, and the AIAG gauge R and R percentage thresholds, express the same underlying idea again at different ratios and for different purposes. So they are related rather than independent. They are still not interchangeable, and quoting one to satisfy a requirement written against another is an argument you lose in an audit.

ISO/IEC 17025 sets no numeric ratio at all. It requires that a decision rule be documented and applied consistently. The default rule in ISO 14253-1, the GPS standard on decision rules, subtracts measurement uncertainty from the specification zone, so conformity is only proven inside a narrowed conformance zone, with the supplier proving conformity and the customer proving nonconformity. If your acceptance practice has never been written down, that default is not what you have been doing.

One last trap, because it is so common. A gauge repeatability and reproducibility study estimates precision only. It does not include the uncertainty of the reference master and it does not detect bias. An instrument can pass a gauge R and R study comfortably and still have a test uncertainty ratio too poor to judge the tolerance you are pointing it at.

• Check your TUR with the free calculator
• Guard band calculator
• Guard banding and decision rules
• Tolerance stack-up analysis

The mistakes that cost the most

• Picking a fit off a list because it looks about right, instead of working out what the joint has to do.
• Deriving a bearing seat from the general fits system rather than from the bearing catalogue, which gives you more interference than you intended.
• Treating a transition fit as though it will land clearance about half the time. It will land where your processes run, and both common machining biases push the same way.
• Adding interference to be safe, when the hoop stress at the hub bore is already higher than the contact pressure you are trying to raise.
• Choosing a bearing fit without deciding the internal clearance class at the same time.
• Specifying a fit for a joint where a key or bolts carry the load and only concentricity is needed.
• Ignoring where in the process the dimension applies, then discovering that plating moved it.
• Tightening a grade on the drawing without asking whether the gauge on the floor can still judge it.
• Assuming a passed gauge R and R study means the measurement is good enough for the tolerance.

Why there are no tolerance values on this page, and where to get them

The limit and fit tables are published in ISO 286-1 and ISO 286-2, which are copyrighted documents sold by ISO and by national standards bodies. We are not entitled to reprint them and we will not. That applies just as much to a calculator that outputs the same values as it does to a table, which is why you will not find one of those here either.

There is a second reason, and it is the one that matters more to you. The published table is the authority. Values generated from the tolerance formulas do not reliably land on the published ones, so a page that quietly computes them is offering you something that looks authoritative and is not. If a number is going on a drawing, it should come from the source.

So get the values from a licensed copy. Buy the two parts of the standard from ISO or your national standards body, use a licensed engineering handbook, or use the tables built into your CAD or PLM system if your licence includes them. Any of those gives you a number you can defend in a supplier dispute. A random web table does not.

Where Axiospec fits

Axiospec does not choose fits and it does not hold the tolerance tables. It is calibration management software, and everything above is free reference content with no account behind it.

The connection is the gauge. Look back at the two facts in the verification section. A limit gauge wears in the direction that makes it accept more, and a limit gauge produces no measured value at all. Put those together and there is no amount of production data that can tell you a GO plug has gone soft. The calibration record is the only mechanism that can catch it, and the finer the grade the less wear it takes to matter.

That is the side Axiospec keeps. Calibration status on every gauge, as-found and as-left readings on every event, and per-instrument uncertainty budgets that feed a test uncertainty ratio so you can see which tolerances your gauges are actually entitled to judge.

The as-found reading is the one that earns its keep here. When a plug gauge comes back out of tolerance, the question is not what to do with the gauge. It is which parts you passed with it since the last calibration, and whether any of them shipped. That is a records question, and it is only answerable if the records were kept properly before anyone knew they would be needed.

• Measurement uncertainty budget calculator
• As-found and as-left readings
• Calibration vs verification