Accuracy vs Repeatability vs Resolution

Accuracy, repeatability and resolution are three separate measures of how well a machine tool — or any measuring device — does what it is asked. Accuracy is how close it gets to the true, commanded position. Repeatability is how close it comes to the same spot every time it tries, whether or not that spot is correct. Resolution is the size of the smallest step it can command or detect. The trio is independent: a machine can be outstanding at one and poor at another. Accuracy vs repeatability is the pairing people most often blur, and resolution is the third term that completes the picture — knowing which one a spec sheet is quoting often decides whether a process holds tolerance.

The three ideas, precisely

Accuracy is the closeness of a measured or achieved position to the target value — the “true” position, established against a known standard. In machine tools it is called positioning accuracy, and its errors tend to be systematic: a machine that always lands a little short of where it is told, in a consistent way across its travel, has a repeatable, mappable error and poor accuracy. Because the error is systematic, it can be measured and largely corrected.

Repeatability is the scatter of repeated attempts to reach the same position under the same conditions. It is a random error, and it does not care whether the position is correct: a machine can be perfectly repeatable while being inaccurate, faithfully landing on the same wrong spot every time. Backlash is the classic repeatability thief when a point is approached from both directions, because the axis takes up its clearance differently each way.

Resolution is the smallest increment of motion the system can command, and the smallest change it can resolve through its feedback. It is the fineness of the grid the machine aims with. Resolution says nothing about accuracy or repeatability: a ruler engraved with half-millimetre divisions has fine resolution, but if the scale printed on it is short, its measurements are still wrong. On a CNC axis, resolution is limited by the weakest link in the chain — if the encoder cannot resolve the steps the controller commands, the encoder decides.

A picture that keeps them apart

The usual picture is a target. Accuracy is how close the shots land to the bullseye; repeatability is how tightly they cluster, no matter where the cluster sits; resolution is how finely the target is ruled. A cluster tight in the wrong corner is repeatable but inaccurate. A cluster scattered around the bullseye is accurate on average but not repeatable. A finely ruled target fixes neither.

Two practical consequences follow. First, repeatability is the foundation: an error you cannot reproduce is an error you cannot measure, calibrate or compensate away — so nothing can be made more accurate than its repeatability allows. Second, resolution is a floor: a system cannot meaningfully deliver better than its smallest step, but a fine step delivers nothing by itself.

The same trio on a CNC axis

Ask a CNC axis to move to a coordinate and a string of things decides how honestly it gets there. The ball screw’s lead error is systematic, so it shows up in accuracy. Backlash and reversal error show up when the axis changes direction, degrading repeatability for points approached from alternate sides while leaving single-direction moves looking fine. Friction and stick-slip add random scatter. Heat is the slow one: the spindle warms, the structure and the screw grow, and positions that were right at the cold start drift as the machine reaches temperature — so accuracy and repeatability are both moving targets during a warm-up cycle.

None of this is accepted passively. CNC controllers carry compensation tables that correct measured pitch error and backlash, and increasingly thermal error as well. The catch is written in the definitions above: compensation only works on errors that are systematic and repeatable. Random scatter, by definition, cannot be mapped. This is why the repeatability figure, not the accuracy figure, is the one that ultimately bounds what a machine can be trusted to do after its offsets are set.

Feedback architecture matters too: measuring the motor or screw (semi-closed loop) cannot see lead error or thermal growth in the axis itself, while a linear scale on the axis sees where the table actually is — and system resolution is set by the least capable element.

Which one should a shop actually care about?

Most production CNC work leans on repeatability. The machinist sets a work offset, or lets a probe measure the part, at the start of the job; the process then assumes the machine returns to those positions part after part and across tool changes. A systematic accuracy error is absorbed every setup by the offset or the probe — which is why a machine with fine repeatability can hold tight parts even when its absolute accuracy is mediocre.

Accuracy steps to the front when position must be right in an absolute sense: features that have to land relative to drawing coordinates shared across several setups, or parts made on more than one machine that must mate or align. No offset rescues a feature when the machine does not know where its own world is. This is when calibration and the published positioning accuracy matter.

Resolution is the lower bound and the source of much marketing. A control that lets you type a fifth decimal is impressive, but mechanics and feedback decide whether the resolution means anything. Read resolution as a ceiling on fineness, never as a promise of accuracy.

One caution when comparing machines: the numbers only mean something if you know how they were measured. Positioning accuracy and repeatability are normally quoted under the ISO 230-2 test code — measured with a laser interferometer, at a controlled 20 °C, through repeated bidirectional moves, reported as statistical figures. Other national standards can produce numbers that look better on paper than ISO 230-2 for the same machine, and any figure is quoted for a cold, unloaded axis. A used machine’s real story is told by measuring it — a laser or a ballbar circularity test — not by its nameplate.

Common misconceptions

• “Fine resolution means it’s accurate.” It means the machine can command small steps; whether it lands them correctly is a separate question.
• “It’s repeatable, so it must be accurate.” No — repeatable-but-wrong is the most fixable state (offset or calibrate it); repeatable is the necessary foundation, not the whole achievement.
• “It approaches the same point fine, so no backlash.” Approaching from one direction only hides reversal error; the bidirectional test is what exposes it.

Accuracy vs repeatability vs resolution is the vocabulary for the question that matters before a job runs: can this machine put the tool where the program says, hold it there part after part, and command the steps the part needs? It is the quality side of the story whose quantity side is told by feeds and speeds — and the whole process runs on the CNC machine that must deliver on both.