Gage R&R (MSA) Calculator

What Gage R&R tells you

Every number you record from a gage is a sum of two things: the true value of the part and the error the measurement system adds. Gage R&R, gage repeatability and reproducibility, is the study that separates those two so you know how much of the variation in your data is real and how much is measurement noise.

It matters because every quality decision you make rests on measurements. If a control chart signals, a capability index looks poor, or a part is scrapped, you need to know whether the process actually moved or whether the gage simply cannot measure consistently.

A measurement system that adds too much error can hide real problems and manufacture false ones, and until you have run a Gage R&R you do not know which world you are in.

The study splits measurement error into two named sources. Repeatability, also called equipment variation, is the inconsistency you get when one operator measures the same part several times with the same gage: it is the gage own inability to give the same answer twice.

Reproducibility, also called appraiser variation, is the extra inconsistency introduced when different operators measure the same parts: it reflects differences in technique, setup, or interpretation. Combined, these give the gage R&R, the total measurement variation, which is then compared against either the total variation in the study or the tolerance to produce a percentage.

That percentage, the %GRR, is the headline result, and the AIAG guideline turns it into a verdict: under 10 percent is acceptable, 10 to 30 percent is marginal, and over 30 percent means the measurement system needs work.

This calculator runs the AIAG average and range method, the standard crossed Gage R&R analysis.

Paste your study data, one line per trial with the operator label followed by one reading per part, optionally add the tolerance, and it returns repeatability, reproducibility, gage R&R, part variation, and total variation, each as an absolute figure and as a percentage, plus the number of distinct categories the system can resolve.

It draws a bar chart of the variance components so you can see at a glance whether the equipment or the operators are the larger problem, and it applies the AIAG thresholds so you get a clear accept, marginal, or reject reading. Everything runs in your browser, and nothing you paste is stored.

How this calculator works, step by step

Begin with a proper study. Select parts that span the normal range of the process, roughly 10 of them, because the study compares measurement error against the spread of real parts, and artificially uniform parts would understate the part variation and inflate the %GRR misleadingly.

Have each operator, ideally 3, measure every part 2 or 3 times, in random order and blind to earlier readings so memory does not bias the repeats. Record the results as one line per trial: the operator label, then the reading for part 1, part 2, and so on across the row. Repeat the operator label on each of that operator trial lines.

The calculator loads with the classic AIAG example, 10 parts, 3 operators, 3 trials, so you can see the format and a real result immediately.

The engine parses your data into operators, parts, and trials, then applies the average and range method. Repeatability comes from the average range of the repeated readings, the typical spread when the same part is measured again, divided by the d2 factor for the number of trials.

Reproducibility comes from the spread of the operator averages, the range between the highest and lowest operator mean, divided by the d2 factor for the number of operators, with the repeatability contribution subtracted out so the two do not double-count. Part variation comes from the range of the part averages divided by the d2 factor for the number of parts.

Each is multiplied by the study-variation constant of 6 to express it as a spread rather than a standard deviation.

The result panel headlines the %GRR against the AIAG thresholds and lists every component in absolute and percentage form. The percentages are the ratio of each component study variation to the total variation, and because the constant of 6 multiplies every component it cancels in those ratios, so the percentages are the same regardless of the multiplier.

The number of distinct categories, ndc, is 1.41 times the ratio of part variation to gage R&R, truncated to a whole number, and tells you how many groups of parts the system can reliably resolve. Enter a tolerance and the tool adds the %GRR of tolerance, comparing the measurement error to the specification width rather than to the process spread.

The bar chart shows %EV, %AV, %GRR, and %PV side by side, coloring the gage R&R bar green, amber, or red by the verdict.

The formulas behind the method

The average and range method estimates each source of variation from a range and a bias-correction factor called d2, which converts an average range into a standard deviation for a given subgroup size. Repeatability variation, the equipment variation EV, is the average range of the repeated measurements divided by d2 for the number of trials, times the study constant.

Reproducibility, the appraiser variation AV, starts from the range of the operator averages divided by d2 for the number of operators, but because that operator range also contains some repeatability, the repeatability variance divided by the number of parts times trials is subtracted before taking the square root, giving the pure operator effect.

Gage R&R is then the square root of the sum of the repeatability and reproducibility variances, because variances, not standard deviations, add.

Part variation PV is the range of the part averages divided by d2 for the number of parts, times the constant, and total variation TV is the square root of the sum of the gage R&R variance and the part variance.

Each component percentage is that component divided by the total variation, expressed as a percent; note that these percentages are ratios of standard deviations, so %EV, %AV, %GRR, and %PV do not add to 100, whereas the underlying variances do. The number of distinct categories is 1.41, which is the square root of 2, times the ratio of part variation to gage R&R, truncated.

When you supply a tolerance, the %GRR of tolerance is simply the gage R&R divided by the tolerance width, comparing the measurement spread to the specification rather than to the observed process spread.

Five worked examples you can follow

Example 1: the classic AIAG study

The calculator default is the standard AIAG example: 10 parts, 3 operators, 3 trials. It returns a repeatability of about 1.21, a reproducibility of about 1.56, and a gage R&R of about 1.98 against a part variation of about 6.84 and a total variation of about 7.12. The %GRR is 27.7 percent, %EV is 17 percent, %AV is 21.9 percent, and %PV is 96.1 percent, with 4 distinct categories. A %GRR of 27.7 percent falls in the marginal band, so this measurement system would be accepted only after weighing the application importance and the cost of improvement; the ndc of 4, just below the desired 5, reinforces that the system is borderline.

Example 2: an acceptable gage

Suppose a study of 10 machined shafts by 3 operators over 2 trials yields a gage R&R of 0.8 against a total variation of 10.5, giving a %GRR of about 7.6 percent. That is under 10 percent, so the measurement system is acceptable and can be used with confidence for process control and capability analysis. With a part variation near 10.5 and gage R&R of 0.8, ndc is 1.41 times 13.1, or about 18, far above 5, meaning the gage can resolve many distinct groups of parts. This is the profile you want: measurement noise is a small fraction of the real part-to-part spread.

Example 3: repeatability dominates

Imagine a %GRR of 34 percent where %EV is 31 percent and %AV is only 12 percent. The system is unacceptable, and the diagnosis is clear from the split: repeatability, the equipment variation, is by far the larger source, so the gage itself is inconsistent. The operators are not the problem. The fix is in the instrument or the fixture, better clamping, calibration, maintenance, or a more capable gage, not in operator training. Re-running the study after improving the fixture would tell you whether the change worked before you trust the gage again.

Example 4: reproducibility dominates

Now imagine the reverse: a %GRR of 28 percent where %AV is 26 percent and %EV is only 10 percent. The gage repeats well, but the operators disagree, so reproducibility dominates. This points to the people and the method: operators may be reading the instrument differently, setting up the part inconsistently, or interpreting an ambiguous characteristic in their own ways. The remedy is a clearer operational definition, standardized setup, training to a common technique, or fixturing that removes judgment. The same overall %GRR can demand opposite fixes, which is exactly why the study separates the two sources.

Example 5: comparing to tolerance instead of total variation

Sometimes the goal is not process control but checking the gage against the spec. Take the same gage R&R of 1.98 from Example 1 and compare it to a tolerance width of 40 rather than to the observed total variation. Using the AIAG 6-sigma study spread, the gage R&R spread is 6 × 1.98 = 11.88, so %GRR to tolerance = 11.88 ÷ 40 = 29.7 percent — a little higher than the 27.7 percent measured against total variation. Which base you choose changes the verdict, so always state whether a %GRR is compared to total variation or to tolerance.

Three expert tips for a trustworthy study

Reading the %GRR verdict

The %GRR is the number most people look at first, and the AIAG guideline gives it three bands. Under 10 percent, the measurement system is acceptable: measurement error is a small enough fraction of the variation that you can trust the gage for process control, capability studies, and part disposition.

Between 10 and 30 percent, the system is marginal, acceptable only depending on the importance of the application, the cost of the gage, and the cost or feasibility of improving it; a marginal gage measuring a critical characteristic should be improved, while a marginal gage on a non-critical feature might be tolerated.

Over 30 percent, the system is unacceptable and must be improved before its measurements can be relied upon, because more than a third of the observed variation is noise the gage is adding.

Two cautions keep the verdict honest. First, the percentage depends on what it is compared against. Compared to total variation, it answers whether the gage can distinguish one part from another, the right question for process analysis. Compared to tolerance, it answers whether the gage can sort conforming from nonconforming parts, the right question for inspection.

A gage can pass one test and fail the other, so choose the comparison that matches the gage job, and this calculator reports both when you supply a tolerance. Second, the bands are guidelines, not physical laws.

A 31 percent result on an inexpensive gage measuring a tight, critical feature clearly needs action; a 9 percent result is comfortable but not a license to stop watching the measurement system, which can drift over time and should be re-studied periodically.

Repeatability versus reproducibility: fixing the right thing

The reason Gage R&R separates measurement error into two sources is that they have different causes and different fixes, and a single overall number would hide which one is at fault. Repeatability, the equipment variation, lives in the gage and the fixture.

When it dominates, the same operator measuring the same part cannot get the same answer twice, and the causes are physical: a worn or dirty instrument, poor resolution, an unstable fixture that lets the part shift, a characteristic that is genuinely hard to locate, or environmental effects like temperature.

The fixes are equally physical, maintenance, calibration, better clamping and location, higher-resolution instruments, or controlling the environment, and none of them involve the operators, so training would waste effort.

Reproducibility, the appraiser variation, lives in the people and the procedure. When it dominates, the gage repeats fine but different operators get systematically different answers, and the causes are human and procedural: operators reading an analog scale differently, setting up or holding the part in their own ways, applying inconsistent pressure, or interpreting an ambiguous characteristic without a shared definition.

The fixes target the method and the people, a precise operational definition of what and how to measure, standardized setup and fixturing that removes judgment, training to a common technique, and sometimes automating the reading so human variation is designed out.

Because the two sources demand opposite responses, always read the %EV and %AV split before acting: the same failing %GRR can mean replace the gage in one study and retrain the operators in another, and the split is the only thing that tells you which.

The number of distinct categories

The %GRR tells you the fraction of variation that is measurement noise, but the number of distinct categories, ndc, answers a complementary and very practical question: how finely can this measurement system actually resolve differences between parts? It is the number of non-overlapping groups of parts the system can reliably tell apart across the observed range of part variation, computed as 1.41 times the ratio of part variation to gage R&R and truncated to a whole number. Intuitively, if the measurement error is large relative to the part spread, the system can only coarsely bin parts; if the error is small, it can distinguish many fine gradations.

The guideline is that ndc should be at least 5. A system with 5 or more distinct categories can be used for variable process control and analysis, because it can resolve enough gradations to detect shifts, estimate capability, and support the arithmetic of control charts. When ndc drops to 2, 3, or 4, the system is losing resolution, and at ndc of 1 it can essentially only sort parts into two groups, functioning as a go/no-go gage rather than a variable measurement.

That has a concrete consequence: a system with low ndc cannot support the statistical tools that assume a continuous measurement, so a capability study or a control chart built on it is unreliable.

Because ndc is driven by the same ratio as %GRR, the two usually agree, a system under 10 percent %GRR will comfortably exceed 5 categories, and one over 30 percent will fall short, but ndc is worth reading in its own right because it frames the gage limitation in terms of what it can and cannot resolve, which is often more intuitive to a shop-floor audience than a percentage.

Total variation or tolerance: which comparison to use

A Gage R&R percentage is a ratio, and what sits in the denominator changes the question you are answering, so choosing the right comparison is essential. Comparing the gage R&R to the total variation, the process spread observed in the study, asks whether the measurement system can distinguish one part from another as they actually vary in production.

This is the comparison you want when the gage will be used for process analysis: for control charts, for capability studies, for detecting whether the process has shifted.

If measurement noise is a large fraction of the real part-to-part variation, the gage blurs the very differences those tools are trying to detect, so the %GRR of total variation is the governing number.

Comparing the gage R&R to the tolerance, the specification width USL minus LSL, asks a different question: can the measurement system reliably sort conforming parts from nonconforming ones? This is the comparison you want when the gage primary job is inspection, deciding pass or fail against a specification.

Here the process spread is irrelevant; what matters is whether the measurement error is small relative to the specification band, because a gage whose error is a large fraction of the tolerance will misclassify parts near the limits, passing bad ones and failing good ones. The two comparisons can disagree sharply. A very capable process with a wide tolerance might have a gage that looks poor against the tight total variation but excellent against the generous tolerance, or vice versa for a marginal process with a snug specification.

This calculator reports the %GRR of total variation by default, and when you enter the tolerance it adds the %GRR of tolerance, so you can judge the gage against whichever standard, or both, matches how you actually use it.

Designing a study that gives an honest answer

The validity of a Gage R&R rests entirely on how the study is run, and a few design choices separate a meaningful result from a misleading one. The most important is part selection. The parts must span the normal range of process variation, because the study measures gage error against the part-to-part spread, and if you hand the operators ten nearly identical parts the part variation collapses, the %GRR inflates, and a perfectly good gage fails the study for the wrong reason. Deliberately choose parts across the range you actually produce, from near one specification limit to near the other, so the denominator of the percentage reflects real production.

Randomization and blinding come next. Each operator should measure the parts in a random, independent order on each trial, and should not see their own earlier readings or those of other operators. If operators measure parts in the same order every time, or can recall what they read a moment ago, repeatability looks artificially small because they are reproducing a remembered number rather than genuinely re-measuring, and the study understates the true equipment variation.

Balance matters too: every operator should measure every part the same number of times, because the average and range arithmetic assumes a balanced crossed design. Finally, run the study under normal conditions with the operators who actually do the job, using the real gage and fixture in the real environment, so the measurement variation you estimate is the one you will live with, not an idealized version produced under special care.

A study run carelessly does not just give a wrong number, it gives a confident wrong number, which is worse than none.

Common mistakes in Gage R&R studies

These errors recur and quietly invalidate measurement studies. Watch for them.

• Uniform parts. Selecting parts that are too similar shrinks part variation and inflates %GRR, failing a good gage. Span the real process range.
• Operators seeing prior results. If appraisers remember or copy earlier readings, repeatability is falsely low. Randomize order and blind the trials.
• Reading only the %GRR. The overall number does not say whether to fix the gage or train the people. Always read the %EV versus %AV split.
• Wrong denominator. Judging an inspection gage against total variation, or a process gage against tolerance, answers the wrong question. Match the comparison to the gage use.
• Confusing precision with accuracy. A low %GRR means precise, not correct. Bias, linearity, and stability are separate studies that need reference standards.
• Too few parts, operators, or trials. A study with a handful of measurements gives an imprecise estimate. Use the recommended 10 parts, 3 operators, 2 to 3 trials.
• Studying once and forgetting. Measurement systems drift. Re-run the study periodically and after any change to the gage, fixture, or method.

Where Gage R&R fits in the quality toolkit

Gage R&R is the foundation beneath every other quality calculation, because all of them consume measurements and none of them can be trusted further than the measurement system that fed them.

This is not a figure of speech: a capability index computed from data laced with measurement noise will understate the true capability, because the observed spread includes the gage error; a control chart built on a noisy gage will either miss real shifts or signal on measurement variation; and a defect count depends on measuring the characteristic correctly in the first place.

Running the Gage R&R first, and fixing the measurement system if it fails, is what makes the downstream numbers meaningful.

Concretely, once the measurement system is validated here, the process capability calculator can turn trustworthy data into Cp, Cpk, Pp, and Ppk, and the control chart calculator can set limits that respond to the process rather than to gage noise.

The defect and yield tools, the sigma level and DPMO calculator and the process yield calculator, likewise depend on measurements that classify parts correctly, and even the acceptance sampling calculator assumes each inspected unit is judged defective or good reliably, which is a measurement question.

In a Six Sigma project, measurement systems analysis lives in the Measure phase precisely because you must confirm the data is sound before analyzing it. Start here, then move to the rest of the Quality Control hub with confidence that your numbers mean what they say.

Precision is not accuracy: the limits of this study

It is essential to understand what a Gage R&R does not measure, because a low %GRR can lull you into trusting a gage that is quietly wrong. This study quantifies precision, the repeatability and reproducibility of the measurement, how consistently the system gives the same answer. It says nothing about accuracy, whether that consistent answer is centered on the true value. A gage can be beautifully precise, every operator getting nearly identical readings, while being biased, reading consistently high or low by a fixed amount, or non-linear, biased by different amounts across its range. Precision and accuracy are independent, and Gage R&R only sees the first.

Accuracy is the domain of three other measurement systems analysis studies, and a complete MSA program includes them alongside the R&R. Bias measures whether the average reading differs from a known reference standard. Linearity measures whether that bias changes across the measurement range, so a gage might be accurate in the middle but off at the extremes. Stability measures whether the bias drifts over time, which is why calibration schedules exist.

All three require a reference standard, a part or master of known true value, which is exactly what a Gage R&R does not need, and that is the practical dividing line: R&R studies variation without a reference, while the accuracy studies compare against one. The takeaway is that passing a Gage R&R is necessary but not sufficient for a trustworthy measurement system. A gage that is precise but biased will consistently misjudge parts, and only a bias or linearity study will reveal it.

Treat this calculator result as the precision half of the picture, and pair it with calibration against traceable standards for the accuracy half.

The business cost of a gage you cannot trust

It is tempting to treat a measurement study as a formality, a box to tick before the real work begins, but an inadequate measurement system carries costs that are large, real, and mostly invisible until you look for them. When too much of the observed spread is measurement noise, two expensive errors follow.

Good parts get rejected, because noise pushes an in-specification reading past the limit, and you scrap or rework product that was fine, paying twice for the noise. Bad parts get accepted, because noise pulls an out-of-specification part back inside the limit, and the escape reaches the customer, where the cost of a field failure dwarfs the cost of the study that would have caught it.

Both errors grow as the measurement error grows, and both are silent, because the operator sees only a number, not the noise inside it.

The subtler cost is corrupted decision-making. A capability study run on a noisy gage understates the true capability, so a capable process looks marginal and gets needless attention, or a marginal one looks capable and escapes it. A control chart on a noisy gage cries wolf, signaling on measurement variation, until operators learn to ignore it, at which point it also misses the real signal.

Improvement projects chase phantom variation that lives in the gage, not the process, wasting engineering effort on a problem that a two-hour measurement study would have located precisely. Set against these costs, a Gage R&R is inexpensive: a morning of measurements and a calculation.

The return is not just a number but the confidence that every downstream decision rests on data that means what it says, which is why measurement systems analysis is treated as a prerequisite rather than an afterthought in any serious quality program.

Data format and quick reference

Enter one line per trial. Start the line with the operator label, a name or letter, then list one reading per part across the row, separated by spaces or commas. Repeat the operator label on each of that operator trial lines, so an operator who measures three times contributes three lines. Every line must have the same number of readings, one per part, and you need at least 2 operators, 2 trials, and 2 parts; the recommended study is 10 parts, 3 operators, and 3 trials. The optional tolerance is the specification width, USL minus LSL, in the same units as your readings. The reference below summarizes the AIAG acceptance guideline.

Frequently asked questions

Sources, disclaimer, and editorial transparency

The method, formulas, d2 factors, and acceptance thresholds used here follow the AIAG Measurement Systems Analysis (MSA) reference manual and the average and range method for a crossed Gage R&R study. This calculator and guide are built and reviewed by the OpsCalculators team; see our Editorial Policy for how each tool is researched, built, and tested.

Results are accurate estimates for planning and education, not certified engineering advice, and this tool implements the average and range method; the ANOVA method additionally estimates the operator-by-part interaction and may differ slightly. Gage R&R measures precision only, not accuracy; pair it with bias, linearity, and stability studies against traceable standards. See our full Disclaimer. OpsCalculators.com is operated by MAFHH INTERNATIONAL LTD. Your inputs are processed in your browser and are never stored; see our Privacy Policy.