What Machinist Tolerances Actually Mean (And Why ±.005 Is a Fight)

May 10, 2026

What Machinist Tolerances Actually Mean (And Why ±.005 Is a Fight)

The print lands on your bench. You scan the title block, the revisions, the GD&T callouts, and then you see it: a basic dimension with ±.0005 written next to it like the engineer was just doodling. You glance at the material — 6061, half a foot long, unsupported on one end. Outside the shop it's 94 degrees. The coolant is the temperature of bathwater. Somewhere a salesperson promised this part by Thursday. This is where tolerance stops being a number on paper and starts being a fight with physics, the machine, the stock, and occasionally the guy who wrote the print.

Tolerance Is Just a Permission Slip

Strip away the geometric symbols and the title-block legalese, and a tolerance is one thing: the range of wrong you're allowed to be before the part stops being a part and becomes scrap. The nominal dimension is what the designer wants. The tolerance is what the designer will accept when reality shows up.

A dimension of 1.000 ±.005 means anywhere from 0.995 to 1.005 is acceptable. The part doesn't have to be 1.000. It probably won't be. Nothing is. Even the gauge blocks you check it with aren't exactly what they claim — they're just closer than whatever you're measuring.

That's the philosophical part. The practical part is that every digit you add to the right of that decimal point multiplies the cost, the cycle time, the operator stress, and the number of times you'll have to walk over to the surface plate before lunch.

The Cost Curve Nobody Shows the Customer

Roughly speaking, here's how the work scales:

**±.010** — A reasonably sharp tool, a halfway-tuned machine, and a guy who knows which end of the calipers is which. Easy money.
**±.005** — Still a manual-machine tolerance if you're careful. On a CNC, it's nothing. On a Bridgeport with a tired leadscrew, it's a conversation.
**±.001** — You're checking with mics, not calipers. You're thinking about tool wear. You're letting the part sit before measuring.
**±.0005** — Temperature matters. The part has to be the same temperature as the gauge. You're probably grinding, or you're indicating in like the part owes you money.
**±.0001** — Now you're in a climate-controlled room, working with a jig grinder or a lapping plate, and the customer is paying for it whether they know it or not.

Every step down that ladder isn't a linear jump. It's a step change in what tools you can use, what processes you can trust, and how many parts you have to make to get five good ones.

Why ±.005 Is a Fight (And It Really Is)

On paper, five thousandths of an inch sounds generous. It's about the thickness of two sheets of office paper. A human hair is around .003. You could feel .005 with a fingernail dragged across a step.

But ±.005 is also where a lot of unexpected things start to matter, and where younger machinists get blindsided because the number looks easy.

Thermal Expansion Doesn't Care About Your Feelings

Aluminum expands about .0000124 inches per inch per degree Fahrenheit. Take a 12-inch piece of 6061, raise its temperature 30 degrees from the morning chill to a hot afternoon spindle, and you've moved roughly .0045 inches. You just consumed almost your entire ±.005 window with weather.

This is why a part measured at the machine, hot off the cut, will read differently than the same part measured an hour later on the surface plate. The QC guy isn't being a jerk. The part actually moved.

Tool Deflection Is Real and It's Local

A 3/8 endmill hanging out an inch and a half from the holder will deflect under a normal cut. How much? Depends on the material, the feed, the rake, the wear, whether you climb or conventional, and whether the spindle bearings have any miles left on them. A few tenths is normal. A couple thou is common in a deep pocket. On a hard material with a worn tool, you can chase a wall around a part and never quite catch it.

Workholding Distorts the Part You're Trying to Hold Still

A thin-wall part clamped in a three-jaw chuck isn't round anymore. It's a triangle that thinks it's round. Cut it, release it, and suddenly you've got a part that's .003 out of round when it was .0005 in the chuck. Same goes for vise jaws on a soft aluminum part — you'll see the jaw marks in the surface finish and the dimensional report.

The Print Says Five Thou. The Process Stack Says More.

Real-world tolerance stacks add up. Machine repeatability, tool runout, fixture variation, material variation, measurement uncertainty — each one eats a slice. By the time you've accounted for all of it, your "easy" ±.005 has maybe ±.002 of actual working room. That's the fight.

GD&T: When ±.005 Isn't Even the Tolerance Anymore

Geometric Dimensioning and Tolerancing exists because plus-minus dimensions break down once you start caring about how a feature is wrong, not just how much.

A hole can be exactly .500 diameter and still be unusable if its position is .010 off from where the print wanted it. So instead of ±.005 on the X and Y coordinates, you'll see a position tolerance — often expressed as a circular zone the hole's center has to fall inside. That changes the math, because now you can trade off X error against Y error as long as the total stays inside the zone.

Flatness, perpendicularity, parallelism, runout, true position, profile — each one is its own kind of tolerance, applied to a feature relative to a datum. None of them are negotiable just because the linear dimensions check out.

If you've ever had a part get rejected even though every caliper reading was inside the box, it was probably a GD&T callout you missed. That's not the inspector being difficult. That's the print doing its job.

The Tools Decide the Tolerance Before You Do

Here's an underrated truth: the tolerance you can actually hold is mostly decided before you ever start cutting.

**Calipers** — Realistically good to ±.002 if you're consistent. Don't lie about this. Calipers are for roughing dimensions and quick checks, not for proving anything.
**Micrometers** — Good to ±.0001 in skilled hands, on a clean part, at the right temperature.
**Bore gauges, indicators, snap gauges** — All reliable, all needing to be set against a known standard.
**CMM** — Whatever the machine's accuracy spec is, divide your usable tolerance by four (the gauge R&R rule of thumb: your measurement system should be at least four times more precise than the tolerance you're checking).

If your shop measures everything with $30 calipers from a tool truck, you don't have a ±.001 capability no matter what the spindle does. You have a ±.002 measurement system masquerading as quality control.

Q&A: The Questions That Actually Come Up

Q: The print says ±.005 but the customer "needs it tight." What do I do?

Make them put the number on paper. "Tight" is not a tolerance. If they want ±.001, they need to write ±.001, and then they need to pay for ±.001. Until then, ±.005 means ±.005 and you make it to the middle of that band if you can.

Q: My CNC holds .0002 in the morning and .0008 by afternoon. Is the machine bad?

Probably not. It's thermal growth in the ballscrews and the spindle. Most production shops either run a warm-up cycle, use scale feedback, or schedule tight-tolerance work for after the machine has been running an hour. Some shops also keep the building temperature stable, which costs money but solves the problem.

Q: Is ±.005 hard or easy?

Yes.

Q: How do I explain tolerance to a non-machinist without sounding condescending?

Tell them the part has to fit between two invisible walls, and the walls are the tolerance. The closer the walls, the longer it takes to make the part not touch them. They'll get it or they won't, and either way they'll still want it Thursday.

Q: Why does the engineer specify tighter tolerances than the function actually needs?

Sometimes because they don't know the function tolerances well. Sometimes because the part has to assemble with other parts that stack up. Sometimes because the last guy who specified loose tolerances got bit. Mostly: pick your battles, and if it's truly unnecessary, ask in a written email so there's a record.

The Honest Math of Plus or Minus a Few Thou

The number on the print is the contract. The number you actually hold is the truth. The space between those two is where a machinist earns their keep — by knowing which tolerance is the real one, which one is aspirational, and which one was copy-pasted from a part the engineer made fifteen years ago and never thought about again.

A few thou is a forgiving universe. A few tenths is a religion. Somewhere in the middle is where most parts live, where most fights happen, and where most of the good work gets done without anybody outside the shop ever noticing.