Tolerance, fit, and clearance

In brief

Ask a machinist to cut a shaft exactly 20 millimeters in diameter and hand it back a hundred times, and you will get a hundred slightly different numbers: 19.998, 20.003, 19.997, never precisely 20.000. This is not a failure of skill or equipment; it is a fact about physical matter and physical tools that no amount of care removes. Because perfect dimensions are impossible, every real dimension must come with a stated allowance for how far it can drift and still work, and that allowance, called a tolerance, is a design decision as deliberate as the nominal size itself. Two toleranced parts meeting each other, a shaft in a hole, a pin in a slot, form a fit, and whether that fit has a gap or an overlap decides whether the assembly spins freely, presses together permanently, or fails outright.

The full treatment

Why the number on the drawing is never the number in your hand

Every manufacturing process has some source of variation: a cutting tool wears as it works, a casting shrinks unevenly as it cools, a die deflects under load, a measuring instrument itself has a smallest readable increment. Even holding temperature constant to prevent thermal expansion does not eliminate variation, because the machine's own repeatability has a limit. The practical consequence is that "make this 20 millimeters" is an incomplete instruction. A complete instruction states the nominal value and the width of acceptable deviation around it, for example 20 millimeters plus 0.02 or minus 0.01. That width is the tolerance. Specifying a tolerance is not admitting defeat; it is converting an impossible demand, exact size, into an achievable one, size within a stated band, and doing so on purpose rather than by accident.

The trade-off that sets the width of the band

A tighter tolerance is not free. Holding a dimension to plus or minus 0.005 millimeters typically demands slower cutting speeds, sharper and more frequently replaced tooling, tighter temperature control, and more inspection, all of which raise cost per part, often steeply as the band narrows below what a process can comfortably hold. A looser tolerance is cheaper to produce but risks parts that do not perform their function: a bearing bore too large lets the bearing rattle, a bore too small will not accept it at all. The engineering task is therefore to assign the loosest tolerance that still guarantees the part's function, no tighter. This mirrors a safety factor: both are a deliberate allowance sized against an uncertainty that cannot be eliminated, one for load, one for dimension, and both cost something if set larger than the function requires.

Fit: what happens when two toleranced parts meet

A fit describes the relationship between two mating dimensions, most simply a shaft and the hole it enters. Define the shaft's actual size as somewhere in its tolerance band and the hole's actual size as somewhere in its own band. Three outcomes are possible depending on how the bands are set relative to each other. A clearance fit places the shaft's whole range below the hole's whole range, so the shaft is always smaller than the hole, guaranteeing a gap; this is used where parts must rotate or slide, such as a shaft in a plain bearing. An interference fit places the shaft's whole range above the hole's, so the shaft is always larger than the hole, guaranteeing the two must be forced together; this is used where a permanent, rigid, immovable joint is wanted, such as a bearing race pressed into a housing. A transition fit lets the two ranges overlap, so depending on which actual sizes come off the production line, a given pair might have a small clearance or a small interference; this is used where alignment matters more than either free rotation or a permanent press, such as a dowel pin locating two plates.

Reading the shorthand and stacking tolerances

Standard systems (ISO and ANSI both use versions of this) write a fit as a nominal size with a letter and number code for each part, for example a hole coded H7 and a shaft coded g6. The letter fixes where the tolerance band sits relative to the nominal size, and the number fixes how wide the band is (a smaller number means a tighter band). This lets an engineer specify a proven combination, H7/g6 for a common sliding clearance fit, without recalculating the physics each time. A second practical issue arises when a dimension is the sum of several toleranced parts stacked in a row, a shaft length made of three collars, say. If every part is at its largest permitted size simultaneously, the stack could exceed the assembly's allowed total length even though every individual part passed inspection. Two ways to bound this exist: worst-case stacking, adding every individual tolerance directly, which guarantees the assembly always fits but often demands unnecessarily tight individual tolerances, and statistical stacking, treating each part's deviation as a random variable and combining variances, which allows looser individual tolerances at the cost of a small, calculated probability that an unlucky combination fails.

Lineage

Interchangeable manufacture, the idea that any part from a batch will fit any mating part from another batch without hand-fitting, is usually traced to musket and firearm production in eighteenth and nineteenth century France and the United States, where armories needed replacement parts to work without a gunsmith custom-fitting each one. Eli Whitney's and later Samuel Colt's factories are commonly cited, though the practical achievement of true interchangeability took decades longer than the popular story suggests. The formalization into limit systems, tables of standard hole and shaft tolerance grades, followed as machine tools and gauging became precise enough to make the standard enforceable, culminating in the ISO and ANSI limits and fits standards used across modern industry.

The strongest case for it

Toleranced fits are why a replacement part ordered from a catalog, a bearing, a bolt, a hose fitting, simply works without custom machining, and why assembly lines can build products from parts made in different factories, sometimes different countries, without ever having the mating parts meet before final assembly. The system scales from a single prototype to millions of units precisely because it separates the question "does this part meet its specification" from the question "will this particular pair of parts work together," answering the second question in advance for every legally produced pair. It also gives manufacturing a clear, checkable pass or fail criterion, which is what makes quality control and supplier contracts possible at all.

The strongest case against it

Tolerance tables describe idealized geometry, a perfectly round shaft, a perfectly cylindrical hole, and real parts deviate in shape as well as size: out of roundness, taper, and surface waviness can make two parts that pass simple diameter checks still bind or wobble, which is why complete drawings add form and position tolerances alongside size tolerances. Worst-case stacking, if used everywhere, can force tolerances so tight that a design becomes unbuildable or absurdly expensive, a common misconception is treating every stack as worst case rather than asking whether statistical stacking is justified. Fits also assume stable conditions; temperature swings expand metals at different rates depending on material, so a fit calculated at room temperature can become loose or, worse, dangerously tight at operating temperature, a failure mode ignored by the tolerance table alone.

Where it stands now

Limits and fits are mature, standardized engineering practice with broad consensus behind both the geometric reasoning and the statistical tools for stacking. What remains a live design judgment, not a solved calculation, is exactly how tight to make any given tolerance, since that choice depends on the specific process capability available, the cost budget, and the function at stake, and reasonable engineers routinely disagree on where the line sits for a new design until process data settles the question.

Test yourself

You are designing a hinge pin that must rotate freely inside a bracket hole for the life of the product, without ever being replaceable in the field. State whether you would specify a clearance, transition, or interference fit, and justify the choice from function alone. Then propose a nominal diameter and a tolerance band for both the pin and the hole such that the loosest possible pairing still rotates freely and the tightest possible pairing never binds, and explain what would go wrong if you had instead specified the two parts to the same single tolerance band centered on the same nominal size.