For ordinary milled or turned parts, a precision band of roughly ±0.005–0.025 mm and as-machined Ra around 0.4–1.6 µm is the realistic specification target, with tighter values treated as feature-specific and premium-priced.

Budget a 2–3× cost step moving from standard (±0.125 mm) to precision (±0.025 mm) tolerance, and a further step-change if grinding, lapping or polishing is added for a finer Ra. The sources do not quantify an ultra-precision tier, so the standard-to-precision comparison is the one you can defend.

What tolerance band can you realistically put on the drawing?

Reported CNC tolerances cluster in a precision band of roughly ±0.005–0.025 mm, with ±0.002 mm or tighter described as possible at the tight end16,21,27,33. Casting sits an order of magnitude looser, at ±0.05–0.3 mm, which is why tight functional features are routed to machining rather than left as-cast16.

The sources disagree on whether the tight end of that band is routine or exceptional. One comparison table presents ±0.005–0.025 mm as typical CNC tolerance with ±0.002 mm possible16; aerospace and mission-critical work presents ±0.005 mm and ±0.001 in as premium, feature-specific achievements27,33.

Treat the band as your default and the tight end as a feature-by-feature exception you can justify, not a blanket note across the drawing. A blanket callout also buys the exception everywhere, including on faces that never touch anything.

If you are still choosing between milling and turning for the part, note that the achievable band is reported the same way for both; the routing decision is geometry and cycle time, not tolerance grade.

Which features actually need the tight callout?

Achievable tolerance is bounded by material stability and feature geometry, not only by the machine. PEEK parts can measure in tolerance and then warp or lose flatness hours or days later as internal stresses relax, and thin geometries are the most vulnerable10,29.

Thin walls deflect and chatter, deep narrow pockets need long-reach tooling with poor rigidity, and small internal radii force small tools at slow speeds30. Run these checks before the grade goes on the drawing:

  • ✓Wall thickness: confirm the section can be machined without deflection or chatter at the grade you intend.
  • ✓Pocket depth and aspect ratio: flag deep, narrow cavities that need long-reach tooling.
  • ✓Internal corner radii: check they suit common tool diameters rather than forcing small tools.
  • ✓Material grade: for plastics, confirm whether the filled grade micro-cracks or relaxes after machining.
  • ✓Datum and setup count: count how many orientations the tight features demand, since each setup adds stack-up risk.
  • ✓Stability over time: for PEEK and similar plastics, ask how flatness is confirmed days after machining.

Which Ra callout, and what operation sits behind it?

Surface finish is bought in discrete steps. Each step down in Ra needs a separate operation or finishing pass rather than a parameter tweak, which is why the finish ladder behaves like a series of added processes8,19,21.

As-machined, turning starts smoother than milling: turning is reported at Ra 0.8–6.3 µm against milling at 1.6–6.3 µm8. Below that, grinding reaches Ra 0.2–1.6 µm and lapping Ra 0.01–0.2 µm, so a sub-micron callout is a grinding or lapping operation, not a tighter CNC pass8 16 (supplier-reported figure).

These ranges are general references, not guaranteed limits, and depend on material, geometry, tooling and machine condition8. Match the callout to function rather than to the smoothest number available.

Structural surfaces are quoted at Ra 3.2–12.5 µm, sliding or rotating surfaces at 0.8–1.6 µm, and sealing interfaces at 0.4–0.8 µm8 16 (supplier-reported figure). If your part is a rotating or sliding interface, the surface finish selection logic follows the same function-first order.

What does tightening actually add to the price?

Both cost sources attribute the step to process time and secondary or inspection operations rather than to material. Tighter tolerance demands slower feed rates, reduced depths of cut, multiple finishing passes, faster tool wear and more CMM and manual inspection time, with higher scrap rates21,23.

That is the mechanism behind the multipliers below.

ChangeReported cost effect
Standard ±0.125 mm → precision ±0.025 mm tolerance (supplier-reported figure)2–3× part cost
As-machined Ra 3.2 µm → mirror Ra 0.2 µm4–5×
Fine machined Ra 1.6 µm vs as-machined (supplier-reported figure)+10–20%
Very fine Ra 0.8 µm vs as-machined+20–40%
Ground Ra 0.4 µm vs as-machined+50–100%
Lapped/polished Ra 0.2 µm vs as-machined+100–300%

Which inspection method can prove the tolerance you specified?

Instrument accuracy spans roughly two orders of magnitude, so the quality plan has to match the callout. CMM is positioned for tolerances tighter than ±0.025 mm and for GD&T, FAI and PPAP documentation24.

The sources set no measurement-uncertainty-to-tolerance ratio or gauge R&R criterion, so ask the supplier how they justify the instrument against your feature rather than assuming the table settles it.

Inspection methodStated accuracyTolerance band it is positioned for
CMM±0.001–0.002 mmTighter than ±0.025 mm, GD&T, FAI, PPAP
Micrometer (outside)±0.002–0.005 mmShaft and pin diameters, flat-part thickness
Optical comparator±0.005–0.025 mm2D profile, radii, angles, edge quality
Dial indicator±0.005–0.01 mmRunout, flatness, parallelism
Digital caliper±0.02–0.03 mmFirst-pass and non-critical features
Contact profilometerRa per stylus traceRa on surfaces that tolerate contact
Non-contact opticalRa by laser or optical scanRa on soft or delicate surfaces

Which certification should the supplier hold?

ISO 9001 is the floor most shops quote against. Where traceability, first-article inspection and process-validation records are contractually needed, AS9100D, ISO 13485 and IATF 16949 are described as jointly requiring documented material traceability, process validation records, FAI data and a controlled, auditable system25.

FAI capability is called a baseline AS9100D requirement for new or changed processes, and counterfeit-parts prevention is mandatory under the standard26. In-process inspection is the practical differentiator.

Evaluation should determine whether the supplier inspects at defined checkpoints or relies only on final inspection26. At volume, capability depends on the QMS, SPC and traceability systems rather than on a single first article12.

Expect to pay for the paperwork. One supplier reports 20–50% premiums for mission-critical space and UAV components over commercial rates, attributed to material certifications, process documentation and quality-system requirements33.

Ask for the documentation package to be named at quotation so no certification requirement surfaces after production begins25.

How do tolerance grade and finish affect lead time and order size?

Tight tolerances introduce additional machining passes and inspection bottlenecks, and high-quality surface finishes require additional post-processing32. Those two effects stack on the same part, so a drawing that is tight on both dimensions and finish carries both delays.

The sources give no MOQ, lead-time or setup-amortization data by tolerance grade for CNC, so you cannot budget a schedule from a grade alone.

Claims of no minimum order quantity and 1–2 business-day lead time exist in the market5, but that is one supplier's stated offer for its own service, not a benchmark for a precision part. Ask each quoting shop what lead time its own inspection queue adds at your tightest feature.

General shop, precision specialist, or a different process?

The sources establish which processes reach which bands, not how supplier tiers compare on cost or lead time. Grinding and lapping reach roughness bands that milling and turning do not, so a sub-micron Ra callout moves the part to a different operation rather than a different shop tier8.

Casting sits at ±0.05–0.3 mm and Ra 3–12 µm as-is, so tight features on a casting are machined afterwards16. 3D printing accuracy varies by technology and generally does not match CNC precision; a common approach is to oversize printed areas and post-machine them6.

For the tier decision itself, the sources offer no comparison of a general CNC shop, a precision specialist and an ultra-precision house on capability, cost or lead time. Qualify the specific supplier against your tightest feature and its inspection evidence instead of choosing a tier by label.

Where the sources disagree, and what to confirm

Two disagreements matter for your budget. Keep both numbers visible and confirm the applicable one with the shop quoting your part.

Disputed item (with unit)One source reportsAnother reportsWhat the buyer should do
Routine CNC tolerance band (mm)±0.005–0.025 typical, ±0.002 possible±0.005 mm as aerospace critical-feature achievementAsk the shop which band it holds routinely on your material and feature size
Cost step, standard to precision tolerance2–3× part costNo comparable multiplier published for this stepRequest the quote broken out by tolerance grade for your part
Cost of a mirror finish (Ra 0.2 µm)4–5× standard Ra 3.2 µm+100–300% for lapped/polished vs as-machinedConfirm which finishing operation the shop prices and at what step

Before you request a quote

  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Request a written quote with price basis, MOQ, tooling, payment terms and production lead time.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
  • Ask the supplier for model-specific evidence and record the condition that would change your buying decision.
Sources · 19

Technical references cited for verifiability — not supplier recommendations. Browse the research library.