CNC Machining: An Accurate Guide
Accuracy in cnc machining is not a machine spec. It is a chain: datum choice, tool engagement, thermal drift, then measurement. This cnc machining accurate guide walks through each link for engineers and buyers. Read it to judge which tolerances your part can actually hold.

In this article
- 1
- 2
- 3
- 4
- 5
- 6
What accuracy actually means in cnc machining
Machine accuracy and part accuracy are different things. A machining center may position to a few microns in a warm room with no load, yet the finished part can miss its target by 0.05 mm. The gap comes from the loop: fixture, tool, material, heat, and measurement. Any cnc machining accurate guide has to start there, not at the spec sheet.
Three words get used interchangeably on drawings. Accuracy is how close you land to the nominal dimension. Repeatability is whether the next part lands in the same place. Resolution is the smallest step the control can command. A machine can be repeatable and still inaccurate, and that is the case that ruins a production run.
The practical question is never whether a shop owns a good machine. It is whether the shop can hold your tolerance across your batch, in your material, after heat treat. That is a process question, and it is answered by fixtures, in-process checks and a stable thermal envelope.
So treat the tolerance block on your drawing as a manufacturing instruction, not a wish. If a feature does not need ±0.005 mm, do not call it out. Every tight callout adds a setup, a probe cycle or a slower finishing pass.
- 1AccuracyHow close the cut lands to nominal.
- 2RepeatabilityWhether part 2 lands where part 1 did.
- 3ResolutionSmallest command step the control can issue.
Datum strategy decides more than the machine does
Most out-of-tolerance parts we see are not machined badly. They are measured or fixtured from a different reference than the one used to cut them. Pick three datums, machine from them, and inspect from them. Nothing else works.
A common mistake is using a raw stock face as datum A. Saw-cut bar stock is rarely flat to better than 0.1 mm over 300 mm, and it moves after the first cut releases internal stress. Face it, then establish datums on machined surfaces. On a plate part, that usually means a skim cut on the bottom before anything else.
For a part with two features on opposite faces, keep the number of re-fixturings low. Every re-clamp adds a small positional error, typically 0.01–0.03 mm if the vise is clean and the stops are solid. Five-axis work helps here: one setup, one datum, both faces cut with the table rotating rather than the part being moved.
Datum targets also matter on thin walls. If you clamp a 2 mm wall from the side, it deflects under the vise jaw and springs back after the cut. The hole measures right with the clamp on and wrong once released. Support the wall from behind, or cut it last with light finishing passes.
- 1Face firstSkim the raw face, then set datums on machined geometry.
- 2One setup if possibleEach re-clamp adds 0.01–0.03 mm of position error.
- 3Never clamp thin walls sidewaysDeflection releases after the cut and the feature moves.
Tool engagement, chip load and the surface you get
Surface finish and dimensional accuracy come from the same variables: cutting speed, feed per tooth, radial engagement and tool runout. Push the radial depth too high on a long tool and it deflects. The wall comes out tapered, and no amount of polishing hides the taper.
A rule that holds on aluminium and steel alike: keep the tool as short as the geometry allows. Deflection scales with the cube of the length overhang. Going from 4×D to 6×D overhang roughly triples the bending. If a deep pocket will not allow a short tool, use a smaller diameter at higher spindle speed rather than a long, flexible one.
For finishing passes, a light radial step-over of 5–8 percent of tool diameter produces a much better Ra than a heavy pass at the same feed. That is how we reach Ra 0.8–1.6 μm on a milled face without a separate grinding step. Below that, Ra 0.2–0.8 μm, we usually add a finishing operation or a different process.
Thermal growth is the quiet one. A spindle running for hours grows, and the tool tip moves with it. On a long cycle, a warm-up run and a stable coolant temperature keep the last part as accurate as the first. On short runs, the effect is small enough to ignore.
- 1Short toolsDeflection grows with the cube of overhang length.
- 2Light finishing step-over5–8 percent of tool diameter for Ra 0.8–1.6 μm.
- 3Warm up the spindleLong cycles drift as the head grows.
How the material sets the accuracy ceiling
Aluminium 6061 and 7075 hold tight tolerances comfortably. They cut fast, generate moderate heat, and do not spring back much. Most of our precision work in these alloys stays inside ±0.005 mm without special measures.
Stainless 304 and 316 work-harden. A dull tool rubs instead of cutting, the surface hardens, and the next pass cuts differently. Sharp tools and a feed high enough to stay under the hardened layer are the fix. 17-4PH in the H900 condition is stable; in the annealed condition it moves during and after machining.
Titanium Ti-6Al-4V has a low modulus and a high cutting temperature. Thin sections deflect under cutting force and then relax. Rough, stress-relieve, then finish. Inconel is slower still, and tool wear between the rough and the finish pass can shift the finish dimension if the offsets are not re-checked.
Plastics behave in the opposite way. PEEK and POM expand with heat and cut with a soft surface. Coolant, sharp tools and a final spring pass get the dimension right. Carbon fibre is abrasive and delaminates if the feed is wrong, so the accuracy conversation becomes a tool-life conversation.
- 1AluminiumHolds ±0.005 mm without special handling.
- 2StainlessWork-hardens; keep the feed under the hard layer.
- 3Titanium and InconelRough, relieve stress, then finish.
Measuring what you actually made
Inspection is where an accurate guide earns its name. A caliper on a curved surface tells you almost nothing. A CMM with a defined datum alignment tells you what the part is. Use the drawing datums, not the convenient ones.
Temperature matters more than most people expect. Steel grows about 11 μm per metre per degree Celsius. A 300 mm part measured at 25 °C and cut at 20 °C differs by roughly 0.017 mm from the thermal offset alone. For a ±0.005 mm callout, that is the whole budget. Let the part settle before final inspection.
For production runs, in-process probing beats end-of-line inspection. Measuring a critical bore while the part is still in the fixture lets the control adjust the offset on the next part. For one-offs, a first-article check with a report is usually enough.
We inspect 100 percent of parts before shipment, and reports are available on request. The useful thing for a customer is not the report itself. It is that the same datums used on the drawing were used to cut and to measure.
- 1Use drawing datumsInspection alignment should match the machining setup.
- 2Let parts settleSteel moves about 11 μm per metre per degree Celsius.
- 3Probe in processCritical bores can be checked before the part leaves the fixture.
Which process to pick for the feature you are making
Match the feature to the process before you argue about tolerance.
| Feature or condition | Process to pick | Why it holds accuracy |
|---|---|---|
| Prismatic part, 3 faces open | 3-axis milling | Fewest setups, shortest tool overhang |
| Features on 4 sides | 4-axis with rotary table | One datum, no manual re-clamp |
| Impeller, blade or undercut | 5-axis simultaneous | Tool approaches from any angle in one setup |
| Shaft with flats and a bore | Mill-turn center | Turning and milling share one datum |
| Thin wall under 2 mm | 5-axis, light finishing passes | Fewer clamps, less cutting force |
| Hardened steel above 45 HRC | Rough, harden, then finish | Finishing after heat treat avoids distortion |
| Titanium thin section | Rough, stress relieve, finish | Removes locked-in stress before final cuts |
| PEEK or POM part | Milling with coolant and spring pass | Controls thermal growth and spring-back |
When five-axis is worth it, and when it is not
If your part has features on three or more faces, undercuts, or a curved surface that must stay true to a single datum, five-axis pays for itself. If it is a flat plate with holes, three-axis is faster and cheaper and just as accurate.
Questions engineers ask before releasing a drawing
Can you hold ±0.005 mm on every feature of a part?
No, and no shop can. ±0.005 mm is achievable on specific features with the right geometry, material and setup.
Long bores, thin walls and deep pockets have different limits. We flag which callouts are realistic during DFM review.
Does five-axis machining automatically mean better accuracy?
It means fewer setups, which removes re-clamp error. That is a real gain.
The cut itself still depends on tool rigidity, feed and thermal stability. A poorly supported five-axis cut is worse than a well-fixtured three-axis cut.
Why did my part measure right at the shop and wrong at my end?
Temperature and clamping are the usual causes. A part measured warm, or measured while still held, changes once it cools and is released.
Send us the measurement method and the ambient temperature, and we can usually reproduce the difference.
How do you handle tolerances on heat-treated parts?
We rough with stock left on, send the part for heat treat, then finish. Finishing after treatment removes distortion rather than trying to predict it.
For 17-4PH and 4140 this is standard practice for any tight callout.
What surface finish can come straight off the machine?
As-machined surfaces typically land at Ra 1.6–3.2 μm. A controlled finishing pass reaches Ra 0.8–1.6 μm.
Below that, Ra 0.2–0.8 μm, we add a finishing operation. Tell us the Ra and the measurement method with the RFQ.
Do you need the full 3D model, or is a 2D drawing enough?
Both help. The model defines the geometry; the drawing defines datums, tolerances and finish callouts.
If you only have a model, send the critical dimensions and we will build the drawing notes with you.
Send the drawing, get a manufacturability answer
Upload your model and drawing. We return a quotation and a free DFM analysis within 12 hours, and production can start within 24 hours.
12-hour quote100% inspectionNo minimum order quantityNDA on request