How to Read CNC Machine Tolerances and Accuracy Ratings

What Do the Three Core Specs Actually Measure

When you open a CNC machine datasheet, three numbers dominate the accuracy section. They are accuracy, repeatability, and positioning precision. Most buyers treat them as synonyms. That is a mistake.

Accuracy describes how close the machine’s actual position is to the programmed position. If the controller tells the spindle to move to 100.000 mm and the workpiece is at 100.005 mm, the machine has a positive error of 5 microns at that point. Accuracy is a systematic deviation. It can be mapped and corrected with compensation tables if the machine supports it.

Repeatability measures how consistently the machine returns to the same position after multiple cycles. If a robot or CNC axis moves to the same coordinate ten times, repeatability is the spread of those positions. It does not tell you whether the average position is correct. It tells you whether the machine will produce the same part shape every time.

Positioning precision is the most practical number for sourcing. It combines accuracy and repeatability into a single metric that describes the machine’s ability to stop where told, over a defined travel range, under defined conditions. It is usually quoted as a maximum error value, such as the worst-case deviation from the commanded position.

The difference matters. A machine with excellent repeatability but poor accuracy will produce identical parts that are all wrong. A machine with poor repeatability but good accuracy will produce parts that drift out of shape. For most production work, positioning precision is the number that drives the decision.

How Datasheet Conditions Change the Numbers

The second thing to check is the condition line below the spec. It will usually say “at 20 degrees Celsius,” “after 30 minutes warm-up,” or “under light cutting load.” That condition changes the number.

A machine measured cold at 15 degrees Celsius will often show better positioning precision than one measured hot at 25 degrees Celsius. Thermal expansion of the gantry, ballscrews, and spindle housing changes the geometry as the machine warms up. If the datasheet does not state a warm-up time, assume the number is measured before the machine reaches its operating temperature. That is a conservative reading, but it is not the number you will see on the floor.

Load matters too. A spec measured with the spindle at zero RPM and no cutting force is not the same as a spec measured with the spindle at full speed and the table under cutting load. Vibration and deflection under load can add error that is not visible in a no-load test.

If you are comparing two machines for a job that involves heavy milling or turning, ask the seller to provide the spec under the load you will actually run. If they cannot, treat the no-load number as a lower bound, not a target.

What Is the Difference Between Backlash and Positioning Error

Backlash is the lost motion in the drive system. When an axis reverses direction, the ballscrew or gear train has a small gap between the drive side and the load side. The controller does not know that gap exists. The axis moves, stops, and reverses, and a small distance is lost.

Positioning error is the total deviation from the commanded position. It includes backlash, but also includes mechanical wear, thermal growth, and control lag. A machine with zero backlash still has positioning error. A machine with high backlash has positioning error that gets worse every time the axis reverses direction.

In a 3-axis milling operation, the axis reverses direction many times per part. If the table has 10 microns of backlash and the controller does not compensate, each reversal adds error. The resulting part may show a small step in the surface or a shift in the hole location relative to the datum.

When reading a datasheet, look for a backlash figure if it is listed. If it is not listed, ask. Some manufacturers hide backlash inside the positioning precision number. That is acceptable, but it means the positioning number includes a fixed amount of lost motion that you cannot remove without a compensation table.

For a job that involves frequent axis reversals, such as contouring or drilling, backlash compensation is a must. If the machine does not support it natively, you may need to program it in the post-processor, which adds complexity and maintenance risk.

How to Use Positioning Precision to Back-Calculate Your Part Tolerance

This is the step most buyers skip. The machine’s positioning precision spec does not directly equal the part tolerance you can hold. You have to account for the other error sources in the chain.

The total error on a finished part is the sum of machine error, fixture error, tool error, and workpiece error. The machine’s positioning precision is only one of those terms. If your machine has 8 microns of positioning precision, but your fixture is not clamped consistently, your tool has 5 microns of runout, and your workpiece is not held flat, your part error will be much larger than 8 microns.

A practical rule is to divide the machine’s positioning precision by a safety factor before assigning it to the part tolerance. For a production job with a fixed fixture and a known tool, a factor of 2 to 3 is common. For a job with manual setup and variable tooling, a factor of 4 or more is safer.

Consider a simple worked example. You are buying a 3-axis milling machine for a batch of aluminum brackets. The datasheet lists a positioning precision of 10 microns at 20 degrees Celsius after warm-up. Your fixture is a dedicated V-block with a clamping force of 500 N, and you have verified that the fixture repeatable to within 2 microns over ten setups. Your end mill has a runout of 3 microns. You are cutting 6061 aluminum at a feed rate that causes no visible deflection.

The total error budget is the positioning precision plus fixture error plus tool error. That is 10 plus 2 plus 3, or 15 microns. With a safety factor of 2, your practical part tolerance is about 30 microns. If your drawing calls for a 20 micron tolerance on the bracket, the machine is not the bottleneck. The drawing is. You either need a better machine, a better fixture, or a tighter process control.

If the same drawing called for a 50 micron tolerance, the 10 micron machine is fine. The machine spec is not the limit. The process is.

How to Verify the Spec Before You Buy

The datasheet number is a claim. You need to verify it.

Ask the seller for the test report. A legitimate manufacturer will have a report from a third-party metrology lab or from their own in-house test rig. The report should state the test standard, the environmental conditions, and the travel range over which the measurement was taken.

If the report is not available, ask for a live demo. Bring your part and your fixture to the factory floor. Run the part on the machine. Measure it with your own micrometer or CMM. Do not trust the seller’s measurement.

If you cannot visit the factory, ask for a video of the machine running your specific part, with the measurement taken by a third party. This is harder to get, but it is worth asking.

Also check the warranty. The warranty should cover the accuracy spec for a defined period. If the spec is not covered, you are buying a machine and accepting the risk that it will drift out of spec after six months. For a production machine, that is a real cost.

Common Mistakes in Reading the Specs

The first mistake is comparing a machine rated at 20 degrees Celsius to one rated at ambient temperature. If one is 18 and the other is 25, the numbers are not directly comparable. Ask both sellers for the spec at the same condition.

The second mistake is assuming that a lower number is always better. A machine with 5 microns of positioning precision but no backlash compensation is not necessarily better than one with 10 microns and full compensation. The compensation removes a large portion of the error. The effective error may be lower.

The third mistake is ignoring the travel range. A spec over a 300 mm range is not the same as a spec over a 1000 mm range. The error grows with travel. If your part is 400 mm long, you need the spec over a 500 mm range, not a 300 mm range.

The fourth mistake is not checking the control. The same mechanical platform can have different accuracy depending on the controller. A basic controller with a 16-bit encoder may not resolve the position as finely as a 24-bit encoder with interpolation. The mechanical spec is the same. The effective spec is not.

When you are comparing machines, put the specs in a table. List the positioning precision, the backlash, the travel range, the condition, and the controller type. Do not compare two machines if any of those fields is missing.

Metric	What It Measures	Why It Matters for Sourcing
Positioning Precision	Maximum deviation from commanded position	Primary spec for part tolerance planning
Repeatability	Spread of repeated positions	Predicts part-to-part consistency
Backlash	Lost motion on axis reversal	Affects contouring and drilling accuracy
Warm-up Time	Time to reach thermal equilibrium	Determines when the spec is valid
Travel Range	Distance over which the spec is rated	Must cover the longest feature on your part

How to Document the Spec in Your Procurement File

Once you have the spec, write it down in the way you will use it. Do not just paste the datasheet page into the file. Translate it into a line that your production team can read.

For example: “Machine A: 10 micron positioning precision, 2 micron backlash, rated at 20 C after 30 min warm-up, over 500 mm travel, 24-bit encoder. Practical part tolerance: 30 microns with dedicated fixture.”

That line is useful. It tells the buyer what the machine can do. It tells the operator what to expect. It tells the quality engineer what to check.

Keep the test report with the machine. If the spec drifts after six months, you have the baseline. If the buyer disputes the machine’s condition, you have the evidence.

Final Thought on Tolerance Planning

The machine spec is the starting point, not the finish line. The part tolerance you can hold is set by the whole system. Machine, fixture, tool, workpiece, and operator all contribute. The datasheet gives you one number. Your job is to turn that number into a process plan that matches the drawing.

If you are buying a machine for a tight-tolerance job, spend the time on the spec verification. The cost of a bad machine is not the purchase price. It is the scrap, the rework, and the downtime that follow.