CNC Accuracy vs Repeatability: Two Numbers That Decide Your Yield
Accuracy is whether your parts land on the target number; repeatability is whether they land on the same number every time. A machine can be perfectly repeatable and still scrap every part — and a shop that quotes only "tolerance capability" without separating the two is hiding which problem you will have. Yield is decided by both.
The two words are used interchangeably on shop floors and in sales literature, which is exactly why so many buyers get surprised. A machine with poor accuracy produces parts centered off the drawing value — every part wrong in the same direction. A machine with poor repeatability scatters parts around a center that may be fine — some parts pass, some fail, unpredictably. Understanding the difference tells you which inspection data to trust, which supplier claims to verify, and where your scrap actually comes from.
The Two Numbers, Defined with Numbers
Positioning accuracy describes how close the machine's commanded position is to its actual position — the systematic error, usually specified as a maximum deviation band like ±0.005 mm over the travel. Repeatability describes how closely the machine returns to the same position when commanded repeatedly — the random error, typically specified as a smaller number like ±0.002 mm. A realistic production machine spec pairs a looser accuracy with a tighter repeatability: the machine reliably returns to where it thinks zero is, and the offset between that and true zero is corrected by probing, calibration, or setup.
| Machine spec (typical, modern CNC) | Meaning | Typical value |
|---|---|---|
| Positioning accuracy | Systematic offset from commanded position | ±0.005 to ±0.02 mm over travel |
| Repeatability | Scatter around the achieved position | ±0.002 to ±0.008 mm |
| Resolution / minimum increment | Smallest programmable step | 0.001 mm or finer |
| Thermal drift over a shift | Position change as machine warms | 0.005–0.03 mm on long axes |
| Backlash (leadscrew machines) | Reversal error on direction change | 0.005–0.02 mm if uncompensated |
The table is the whole argument: thermal drift and backlash are usually larger than the repeatability spec, which is why real parts are not machined to the nameplate — they are machined to a process that measures, compensates, and checks. A machine's stated accuracy matters far less than the shop's method for finding and correcting the offset on your specific part.
How the Two Numbers Create Yield
Yield comes from the combination of where the process is centered (accuracy) and how wide it spreads (repeatability). Picture a part with a ±0.05 mm tolerance band. A process that is accurate but not repeatable scatters parts across the band — a fraction falls outside on both sides, and that fraction is your scrap. A process that is repeatable but inaccurate produces a tight cluster of parts sitting outside the band — every part scrap, consistently, until someone notices and shifts the offset. The second failure is more dangerous because it looks like success: the parts are all identical, and identical wrong parts pass an in-process check that only looks at consistency.
Process capability indices capture this. Cpk measures how centered the process is within the tolerance; it drops when the process mean drifts off target. Pp and Pk measure the same on a short-term sample. A shop targeting Cpk above 1.33 — the common industry minimum — needs the process mean within about one eighth of the tolerance band of the target, and the spread narrow enough to fit the remaining width with margin. When you see a first-article report showing every feature dead center, and a production batch showing the same features drifting 0.02 mm off, you are watching accuracy drift — usually thermal — defeating an otherwise repeatable process.
Which One Should You Ask a Supplier About?
Ask both, in separate questions. "What accuracy does your machine hold?" gets a marketing answer. "What is the Cpk on the critical features of this part, and what is the process mean versus the drawing nominal?" gets a real one. The first article tells you the accuracy of the setup — whether the shop found the true position of the part and the tool offsets. The batch report tells you the repeatability — whether the process holds that position across the run, through tool wear and temperature change. Our guide to reading CMM inspection reports shows exactly where to look for the deviation column on both documents.
| Situation | Accuracy problem | Repeatability problem |
|---|---|---|
| First article | All features off in one direction | Scatter exceeds tolerance |
| Production batch | Drift over time (warm-up, tool wear) | Random variation run to run |
| Between batches | Offset changed after setup change | Same part, different results |
| What fixes it | Probing, tool offset correction, calibration | Better fixturing, thermal control, machine condition |
The useful follow-up question for a supplier is about their process control, not their machine brand: do they probe critical features in-cycle, do they verify with a CMM on the first article and at intervals through the batch, and what do they do when the trend shows drift before parts go out of tolerance? In-cycle probing turns a repeatability-limited process into an accuracy-corrected one, because the machine measures the real position and adjusts before the next cut. That is how modern precision CNC work holds ±0.005 mm in production rather than in a brochure.
The Temperature Factor Everyone Forgets
The largest single source of accuracy drift in a real shop is temperature — and it is invisible on a machine spec sheet. A CNC machine grows as it warms: spindles, ball screws, and castings all expand, and a long-axis position can shift tens of microns between a cold start and steady state. The same applies to the part and the inspection room: aluminum moves about 23 µm per meter per degree Celsius, so a 200 mm part measured at 22°C and assembled at 30°C is a different size by about 37 µm — enough to flip a ±0.02 mm fit.
Shops that hold tight tolerance manage temperature deliberately: they let machines warm up before critical work, they machine and measure in the same environment, and they schedule the tightest features for the thermally stable part of the day. When a batch report and a customer's incoming inspection disagree by a few microns, temperature is the first suspect, and the resolution is agreeing on measurement conditions rather than arguing about whose CMM is better.
How This Applies to Threads, Fits, and Real Parts
The accuracy-versus-repeatability lens applies to every feature, but it shows up most painfully on features with one-sided risk. Threads are a good example: a tap that drifts oversize fails the gage, and an undersize thread is a rework. Our threading guide explains how process choice bounds that risk. The same logic governs fits: a bore that is accurate but not repeatable makes every assembly a gamble, while a repeatable but offset bore makes every assembly fail identically — which is at least diagnosable.
For the buyer, the practical checklist is short. Demand a first-article report with measured values against nominal, not just pass/fail stamps. Demand batch inspection on critical features with the actual numbers. Ask what the process mean is versus the drawing nominal — a shop that knows it is holding the process 0.01 mm high on a bore has told you more about their quality system than any certificate. And remember that the machine is only half the story: accuracy and repeatability are properties of the whole process — machine, fixture, tooling, temperature, and inspection — which is why the same model of machine in two different shops produces two different yields.
Email sc@bquq.com or WhatsApp +86 137 1315 7787 with your PDF/DXF/STEP file. An engineer reviews it and replies with price, lead time and DFM notes on working days.
Frequently Asked Questions
Q: What is the difference between CNC accuracy and repeatability?
A: Accuracy is how close the machine gets to the commanded target — systematic error. Repeatability is how consistently it returns to the same position — random error. A machine can repeat perfectly while being inaccurate, producing identical parts that are all off-target.
Q: Which matters more for production yield, accuracy or repeatability?
A: Both, in sequence. Repeatability determines the spread of your parts; accuracy determines whether that spread is centered on the target. A process needs repeatability to be controllable and accuracy to be correct — and Cpk scores both against the tolerance band.
Q: What does Cpk of 1.33 mean on my parts?
A: It means the process spread plus centering leaves enough margin that the expected defect rate is very low — roughly 63 parts per million outside tolerance if the process stays centered. Higher Cpk means more margin against drift; below 1.0 means parts are already falling outside tolerance.
Q: Why do my parts measure differently at the supplier than at my incoming inspection?
A: Temperature is the most common cause — parts, machines, and gages all expand and contract with temperature, and aluminum moves about 23 µm/m per °C. Measurement method and datum setup also differ between labs. Agree on measurement temperature and method with the supplier before comparing numbers.
Q: How can a supplier hold ±0.005 mm if the machine spec says ±0.01 mm?
A: By correcting the process rather than trusting the nameplate: in-cycle probing finds the real position, tool offset compensation corrects systematic error, and CMM verification confirms the result. The effective accuracy of a probed, compensated process is much better than the raw machine spec.
Authored by the BQUQ Engineering Team. BQUQ is an ISO9001-certified source factory in Dongguan, China, running CNC machining, metal stamping, custom springs, heat sink and collet lines under one roof. Send drawings to sc@bquq.com or WhatsApp +86 13713157787 for a quote within 12 working hours. www.bquq.com
