GD&T for Machined Parts: Tolerances Machinists Can Actually Hold

Geometric dimensioning and tolerancing is the difference between a print a machinist can quote in an afternoon and one that generates three rounds of emails before the first chip gets cut. Most of the cost trouble we see on incoming RFQs is not hard geometry — it is tolerances that got copied from a template, applied to the whole part, or tightened "just to be safe." This article is about the handful of GD&T callouts that actually move machining cost and scrap, and how to specify them so the part lands in spec without paying for precision nobody needs.

A tight tolerance is not a good tolerance

The most expensive habit in mechanical design is tightening a tolerance because it feels safer. A feature that locates a plastic grommet does not need ±0.01 mm, but if a print calls it out, the shop has to machine to it, measure it, and document it — and that cost is real even when the function does not care. We have watched the same aluminum housing quote at roughly 18 percent less the moment a designer relaxed a handful of ±0.01 mm callouts back to ±0.05 mm, with no change to how the part performed in the assembly.

Tolerance cost is not linear. A general tolerance of ±0.05 mm on a milled pocket is a normal day. ±0.02 mm starts to require slower finish passes, more frequent tool changes, and in-process checks. ±0.005 mm moves the part into a different workflow altogether — thermal compensation, staged roughing and finishing, and measurement on a CMM rather than with calipers. Each step down the tolerance ladder roughly doubles the touch time on the feature, and touch time is what you pay for. The rule we give customers is simple: call out the loosest tolerance the function can live with, and reserve the tight ones for the two or three features that genuinely mate, seal, or locate.

GD&T exists precisely so you do not have to over-tighten. A flatness or position callout describes what matters — how a surface sits, how a hole is located — instead of asking every feature on the part to be uniformly perfect. The more precisely you say what you need, the less you pay for what you do not.

The symbols that actually drive a machined part

You do not need the whole Y14.5 library to specify a machinable part. In practice, five or six controls cover almost everything that crosses a CNC shop floor, and each one tells the machinist a different thing about what to hold and what to let go.

True position locates a feature — usually a hole or a boss — relative to a datum reference frame. It is the workhorse of machined parts because holes are how parts bolt together, and position is what guarantees the bolt pattern lines up. Profile of a surface controls the shape of a complex or curved face in one callout, and is the cleanest way to control an irregular machined pocket without piling a dozen linear tolerances on top of each other. Flatness says a single face must sit within a tolerance zone independent of any datum — it does not care where the face is, only how wavy it is. Perpendicularity and parallelism relate one surface to another, and runout ties a rotating surface back to an axis, which is why it shows up constantly on turned shafts and bearing seats.

One warning worth stating outright: concentricity and symmetry are expensive controls that many shops now steer designers away from, because a true position callout with a zero tolerance at maximum material condition achieves the same coaxiality with a friendlier inspection path. If your print still uses concentricity, expect a question from the quoting engineer.

Datums first: pick the frame the machinist can hold

Every GD&T control hangs off a datum reference frame, and the datum scheme is where prints quietly go wrong. A datum should be a surface or axis the shop can physically locate from — a face that sits in a fixture, a bore that takes a mandrel, a flat that registers against a stop. When a drawing picks a datum that is hard to reach or that changes between setups, the machinist is forced to transfer the reference, and every transfer stacks error onto the tolerance budget you already spent.

The classic fixture rule is 3-2-1: three points establish a primary plane, two points establish a line on the secondary, one point stops rotation on the tertiary. If your datum scheme mirrors how the part will actually sit in a vise or a chuck, the part becomes cheap and repeatable. If it does not, the shop builds an elaborate fixture to hold a reference frame the function never needed. The best datum on a turned part is usually the bore or the OD it turns around, because that is the axis the lathe already owns. On a milled plate, the primary is the largest flat face, because that is what drops onto the table first.

A useful test before you release a print: hand the drawing to someone and ask them to describe, in one sentence, how they would hold the part to make the first feature. If they hesitate, the datum scheme is doing too much work. Simplify it until the holding answer is obvious.

True position and the bonus tolerance left on the table

One of the least-used and most valuable ideas in GD&T is the material condition modifier. A position callout with an M in a circle — maximum material condition, or MMC — gives the part a bonus tolerance as the feature departs from its maximum material size. A hole at MMC is at its smallest, so it needs the full position zone to stay clear of the mating pin; as the hole is bored larger, it can drift further off location and still assemble, and MMC lets the drawing say so automatically.

The payoff is real. A bolt hole specified at Ø6.0 mm with a position tolerance of 0.1 mm at MMC can gain up to the size tolerance of the hole in additional position allowance once the hole is bored toward its high limit. That extra allowance is often the difference between a part that needs a second operation to correct location and one that passes on the first try. It costs nothing to add the MMC modifier, and it routinely cuts scrap on hole patterns where the functional requirement is only "the bolt goes through and the parts line up."

MMC only makes sense for features of size — holes, slots, bosses — where "maximum material" has a physical meaning. On a flat surface it does not apply, and forcing it there is a drafting error that will confuse the inspector more than it helps the machinist.

Profile controls for complex surfaces

When a pocket or a sealing face is not a simple flat or cylinder, a linear ± tolerance is the wrong tool. A curved sealing groove controlled by a stack of coordinate dimensions ends up over-constrained, because each dimension has its own zone and the corners where they overlap create ambiguity. Profile of a surface replaces that stack with a single tolerance zone of uniform width that wraps the entire feature — the shape must lie inside one band, not inside a dozen separate boxes.

Profile is also how you control a surface that has to look a certain way, not just locate a certain way. A gasket face that must seal against a mating part gets a profile callout because what matters is the form of the whole face, not the coordinate of any single point on it. The same logic applies to a pocket that seats an O-ring: the groove width and depth matter together, and profile says so in one line instead of three.

The trade-off is inspection. A profile zone on a freeform surface is verified on a CMM with a scan, not with a micrometer, so profile on complex geometry carries a higher measurement cost than a simple flatness or position callout. Use it where the form genuinely matters — sealing, mating, mating dynamics — and fall back to simpler controls where a flat and a hole will do.

A tolerance checklist before you release the print

Every item on this list earned its place from a specific job where a drawing issue cost a customer money or a rework cycle. It is the same list we walk a new print through before we quote.

• Find the three features that actually mate. Tighten only those. Let the rest ride on the general tolerance block.
• Check the general tolerance block. A blanket ±0.05 mm title-block note is usually plenty for non-critical features and far cheaper than per-feature callouts.
• Choose datums the shop can hold. The largest flat face as primary, a bore or edge as secondary. If the holding answer is not obvious, simplify.
• Use position, not concentricity, for coaxial features. A zero-at-MMC position callout is cheaper to machine and easier to inspect.
• Add the MMC modifier to features of size. It is free allowance on holes, slots, and bosses. It costs nothing and cuts scrap.
• Use profile for complex faces, not coordinate stacks. One band beats a dozen overlapping ± boxes on curved and sealing surfaces.
• Match surface finish to the function. A 0.8 µm Ra on a non-sealing cosmetic face is money spent for nothing. Tie Ra to where it seals or slides.
• Say what the feature does, not what you are afraid of. "Locates a pin" is a position callout. "Must not leak" is a profile and a flatness callout. Fear is not a tolerance.
• Leave a margin for tooling. A 0.005 mm tolerance that eats the whole machine capability leaves no room for tool wear — and you pay for every tool change.
• Send it to the shop before you lock it. A five-minute DFM read catches over-tolerancing and bad datums before they become a quote and a schedule.

"A customer sent us a housing with ±0.01 mm on every hole and a flatness callout on a face that only sat against a rubber pad. We quoted it as drawn, then quoted it again after a ten-minute call where we walked the print together. The relaxed version came in 22 percent lower and machined a full day faster, and the parts fit the fixture exactly the same. The drawing never got worse — it got more honest."

GD&T is not paperwork; it is the language that tells the shop what to hold and what to leave alone. A print that uses it well is cheaper to quote, faster to machine, and easier to inspect, because the tolerances describe the function instead of the designer's anxiety. Tighten the three features that mate, choose datums the machinist can actually hold, spend the MMC modifier where it applies, and let everything else ride on the general block — and you will get the same part, sooner, for less.

When the print is fully dimensioned, the next step is proving the process can hold it — see how our quality assurance team reports FAI and SPC data.