Gage R&R: Measurement System Analysis Explained

By Linh Ngo · Updated Oct 10, 2026

Gage R&R (gauge repeatability and reproducibility) is a measurement system analysis study that estimates how much of the variation in your data comes from the measuring process itself rather than from the parts being measured. Repeatability is the variation when one person measures the same part repeatedly. Reproducibility is the variation between different people. Together they tell you whether your data can be trusted.

What is a Gage R&R study?

Every number on a quality dashboard is a measurement, and every measurement carries error. A Gage R&R study puts a figure on that error before you spend months improving a process based on it.

The idea is simple. You take a sample of parts that cover the normal range of your process, have several operators measure each one several times, and then split the total variation into pieces. Some of it is real difference between parts. Some of it is the gauge disagreeing with itself. Some of it is operators disagreeing with each other. The NIST/SEMATECH e-Handbook treats this as part of characterizing a measurement process, alongside stability, bias, resolution and linearity.

The two core terms, as JMP defines them:

• Repeatability (equipment variation): the variation you see when the same measurement is made multiple times. It's the "within" noise of the gauge.
• Reproducibility (appraiser variation): the variation that comes from different operators measuring the parts, including any operator-by-part interaction.

Add them up and you get total Gage R&R. Compare that to the total variation in the study, or to your tolerance, and you've got your verdict on the measurement system.

Why it matters in the Measure phase

Gage R&R is a standard checkpoint in the Measure phase of DMAIC, the improvement cycle behind Six Sigma. The logic is blunt: if the gauge is noisy, a capability number, a control chart and a before-and-after comparison are all contaminated. You can't improve what you can't measure, and you can't measure well without checking the ruler first.

A quick scenario. A plant sees 8% scrap on a machined shaft and starts adjusting the lathe. Three weeks later the scrap rate hasn't moved. A Gage R&R study then shows the bore gauge reads differently by shift. Some of the "scrap" was good parts failing a bad measurement, and some bad parts were slipping through. The lathe was never the main problem.

Crossed vs nested Gage R&R

The design you choose depends on whether every operator can measure the same parts.

Crossed design. Every operator measures every part. This is the default for non-destructive tests because a part survives measurement, so it can be handed from person to person. Minitab's overview gives an example layout of 10 parts, 3 operators and 3 measurements per part, in random order, with parts that represent the expected range of process variation.

Nested design. Each operator measures a different set of parts. You need this when testing destroys the part (a tensile test, a crash test, a chemical assay on a consumed sample) and you can't get enough nearly identical specimens to give each operator the same ones. As Minitab's guide to destructive testing puts it, each part is "nested" within an operator because each part is unique to that operator.

The measurement counts come from Minitab's own worked comparison of the two layouts. For a deeper treatment of the nested case, see the Minitab destructive testing guide. If you have extra factors (say, two gauges and three fixtures) or unbalanced data, Minitab points to an expanded study instead.

How to run a Gage R&R study

1. Define what you're measuring and why. Pick one characteristic, one gauge, one procedure. Write down the tolerance if there is one.
2. Choose parts that span the process. Don't grab ten good-looking parts. Include some from the low end and the high end of normal production. A narrow sample makes the gauge look worse than it is, because part-to-part variation shrinks.
3. Pick operators who really use the gauge. Use the people who measure on the floor, not your most careful technician.
4. Number the parts and hide the numbers. Operators shouldn't know which part they're holding or what they measured last time.
5. Randomize the order. Run each operator through the parts in a different random sequence for each round. Don't let them do all three readings of a part back to back.
6. Record every reading. Use the template below so no cell is left blank.
7. Analyze with ANOVA or the Xbar-R method. Most software offers both. ANOVA is the more informative of the two because it can test the operator-by-part interaction.
8. Read the output, decide, and act. Check the percentages, the number of distinct categories and the graphs together (more on that below).
9. Fix and re-run. If the system fails, improve the gauge, the fixture, the procedure or the training, then repeat the study to prove the fix worked.

How to read the results

Three numbers come up in nearly every report.

%Study Variation and %Tolerance

These compare the measurement system's spread to the total spread seen in the study (%Study Var) or to the specification width (%Tolerance). The widely used bands from the AIAG MSA manual, as Minitab's guide to interpreting gage output reports them:

When you have specification limits, %Tolerance often matters more than %Study Var, because it reflects the product's real limits. If Total Gage R&R takes more than 30% of the tolerance, the gauge may not reliably separate good parts from bad.

%Contribution

This is the same decomposition expressed in variances rather than standard deviations, so the percentages look smaller. The same Minitab article lists less than 1% as acceptable, 1% to 9% as dependent on context, and above 9% as unacceptable. If you quote both sets of figures in a report, label which one you mean. People mix them up constantly, and a 9% contribution is not the same thing as a 9% study variation.

Number of distinct categories (ndc)

The ndc tells you how many separate groups of parts your measurement system can reliably tell apart. Minitab calculates it by dividing the part standard deviation by the gage standard deviation, multiplying by 1.41, and truncating the result. The AIAG MSA manual recommends that five or more categories indicates an acceptable measurement system.

A study can pass on %Study Var and still fail on ndc, or the reverse, so look at both.

The graphs

Don't skip them. The components-of-variation chart should show part-to-part as the biggest bar. Minitab's key results guide says the R chart by operator should stay in control, the Xbar chart should show most points outside the limits (the gauge sees real part differences), and the lines in the operator-by-part interaction plot should be roughly parallel. Lines that cross mean operators rank parts differently, which is a reproducibility problem worth chasing.

Gage R&R template

Copy this layout for a crossed study with 10 parts, 3 operators and 3 trials. Randomize the run order and record it in the first column.

And a summary table to fill in once the analysis is done:

Ten parts times three operators times three trials gives 90 readings. That's a lot of clicking, but it's a one-afternoon job for most gauges.

Examples by business function

For go/no-go and category judgments, like the invoice example, you'd use an attribute agreement analysis rather than the variable Gage R&R above. The thinking is the same, the math isn't.

Gage R&R vs related quality tools

Gage R&R is easy to confuse with the tools that consume its data. It's the one that checks the data first.

The practical order is: Gage R&R, then a control chart to check stability, then a capability study, then improvement work such as DOE. Skipping the first step is the classic way to waste the other four.

Do's and don'ts

Do:

• Use parts from across the full process range.
• Let real operators use the real procedure under real conditions.
• Randomize and blind the runs.
• Look at the graphs, not just the summary percentages.
• Re-run the study after any fix, and repeat it on a schedule for critical gauges.

Don't:

• Hand-pick parts that look "typical." You'll hide part variation and punish the gauge.
• Let operators see earlier readings or compare notes.
• Treat 10% to 30% as a free pass. That middle band is a judgment call, so document why you accepted it.
• Test a gauge that is out of calibration. Calibrate first.
• Use a variable-data study for pass/fail judgments.

Key Facts

• Gage R&R splits measurement variation into repeatability (same operator, repeated readings) and reproducibility (different operators).
• Crossed studies have every operator measure every part. Nested studies give each operator different parts and suit destructive testing.
• AIAG bands as reported by Minitab for %Study Var and %Tolerance: under 10% acceptable, 10% to 30% depends on the application, over 30% unacceptable.
• The same source lists %Contribution bands of under 1% acceptable and over 9% unacceptable.
• The AIAG MSA manual recommends five or more distinct categories for an acceptable measurement system.
• Minitab's example layout uses 10 parts, 3 operators and 3 measurements per part in random order.
• Run it in the Measure phase of DMAIC, before control charts and capability studies.

Where to go next

Once your measurement system passes, you can trust the rest of the toolkit. Start with a control chart to check stability, then calculate Cpk. If you're mapping where this fits in a larger improvement effort, DMAIC is the right frame, and Six Sigma gives the wider context.