A buyer sent us a print once where every single dimension carried the same tight callout. Every one, including a clearance hole nobody would ever measure. The quote came back at roughly three times what that part should cost, and the buyer concluded we were expensive. We weren't. The print was, and he had no idea which line on it was doing the damage.
What a tolerance actually is
No manufacturing process produces the same measurement twice in a row. A tolerance is the band around a nominal dimension inside which the part still does its job. Specify a shaft at 20 ±0.05 mm and anything between 19.95 and 20.05 mm passes inspection. Outside that band, it gets rejected.
The band exists because physics does not negotiate. Tools wear, metal grows with heat, material deflects under cutting force. Your job as the engineer is not to eliminate that variation. It is to bound it at whatever the function requires and not one micron tighter. Which leaves the harder question: how does the shop know where you drew that boundary?
Three ways a print says it, and one that says nothing
- Bilateral: 20 ±0.05 mm. The allowance splits evenly above and below nominal.
- Unilateral: 20 +0.00 / −0.05 mm. All of the allowance falls to one side, usually because the part has to enter something.
- Geometric (GD&T): controls form, orientation and position such as flatness, concentricity and true position, not just a linear size.
The fourth case is the one that costs money. When a dimension carries no callout at all, the general tolerance in the title block applies, often something like ISO 2768-m. Reading a blank as "make it as precise as you can" is one of the most reliable ways to inflate a lot, and it happens more often than anyone admits. To see why, it helps to know what each process can genuinely hold.
What each process can actually hold
Every process has a range where it is both accurate and economical. Push past that range and you are not buying precision, you are buying a second operation. These are the capabilities we run on the floor:
| Process | Typical tolerance | Where it belongs |
|---|---|---|
| CNC milling | ±0.025 mm standard, to ±0.01 mm on fits | Profiles, pockets, faces |
| CNC turning | ±0.013 mm on diameters | Shafts, bushings, rotating assemblies |
| Wire EDM | ±0.005 mm | Sharp internal corners, material to HRC 65 |
| Grinding | ±0.002 mm | Sealing faces, guides, flatness |
Notice that the tightest numbers belong to the slowest processes. That relationship is the entire cost story, and it is not linear.
Why one more decimal place is not one more increment of cost
Each step toward tighter control buys you slower finishing passes, better tooling, sometimes a temperature-controlled environment, and inspection of every piece instead of a sample. Going from ±0.1 mm to ±0.01 mm on one feature can double the unit cost of that feature on its own. Do it across forty dimensions and you have not bought a better part. You have bought forty separate cost multipliers, most of which do nothing.
The expensive thing is almost never a decision
Here is what we see in practice. The overspecified prints that arrive here are rarely the product of an engineer weighing function against cost and choosing tight. They are defaults, copied from the last drawing, inherited from a template, or set once in a title block years ago and never revisited. Nobody decided to pay for that clearance hole. Nobody decided anything at all, and that is exactly why it costs so much.
The fix is not tighter engineering, it is a deliberate one. Specify the loosest tolerance that still guarantees function. Reserve the tight callouts for mating surfaces, seals and critical interfaces. Let everything else fall to the general tolerance on purpose, because you chose to, not because nobody looked.
Three checks before you send it out
Before a drawing goes out for quote, confirm that every tight tolerance traces to a real function, that non-critical dimensions genuinely fall under the general block, and that the surface finish is not overspecified out of the same reflex. A well-toleranced print quotes faster, machines cheaper and comes back through inspection with fewer rejections. If you are unsure what is realistic for your geometry, the efficient move is to review it with whoever will cut it, before the first chip.
What is a good tolerance for CNC machining?
The loosest one that still guarantees the part functions. As a practical range, standard milling holds about ±0.025 mm and turning about ±0.013 mm on diameters. Anything tighter should trace to a specific fit, seal or interface, not to habit.
What happens if a drawing has no tolerance on a dimension?
The general tolerance in the title block applies, commonly a standard like ISO 2768-m. It does not mean "as precise as possible". Assuming it does is a frequent and expensive misunderstanding between design and the shop floor.
Why does a tighter tolerance cost so much more?
Because it brings slower finishing passes, higher-grade tooling, sometimes an additional process such as grinding or EDM, and piece-by-piece inspection instead of sampling. Tightening one feature by an order of magnitude can double the cost of that feature.
What is the tightest tolerance you can hold?
About ±0.005 mm on wire EDM and ±0.002 mm on grinding, depending on geometry, material and part size. We will tell you before quoting whether your print is realistic for the process it needs.