How to Run a Gage R&R in Minitab Step by Step

By Hélène · May 12, 2026

Measurement system analysis is the unglamorous foundation of every capability study, every SPC chart, and every supplier qualification. If the gage cannot reliably tell good parts from bad, no amount of process improvement work will produce trustworthy results. Gage R&R quantifies how much of the observed variation comes from the measurement system itself versus the actual parts.

This guide walks through running a crossed Gage R&R study in Minitab step by step. It is written for quality engineers and process engineers who have access to Minitab and need to plan, run, and interpret a study without guessing at the menus.

Before you open Minitab

A Gage R&R is only as good as the study design. Plan the following before collecting any data.

Sample size and design

The standard crossed design uses:

• Ten parts that span the expected range of process variation.
• Three operators (sometimes called appraisers).
• Three trials (each operator measures each part three times).

That gives 90 measurements total. Skimping on parts or trials weakens the analysis. Using only five parts often hides part-to-part variation and inflates the percent contribution of the measurement system.

Part selection

Parts should represent the real process spread, not just the center. If you only sample parts near the nominal, the part variation looks small and the gage looks worse than it really is. Pull parts from across the tolerance range and across recent production lots when possible.

Randomization and blinding

Operators should measure parts in random order across trials, not in the same sequence each time. Ideally, operators do not know which part they are measuring. This prevents memory bias from inflating reproducibility numbers.

Data collection sheet

Build a data collection sheet that records Part ID, Operator, Trial, and Measurement. Keep it simple. If you are running the study in a CNC machining cell measuring shaft diameter, the sheet might look like 90 rows with those four columns and a notes field for anomalies. If you are stratifying defect data alongside this work, our guide on Minitab vs QI Macros for data stratification covers how each tool handles the upstream cleanup.

Step by step in Minitab

Step 1: Stack the data

Minitab expects stacked data, not a wide format. Three columns:

• Part - part identifier (1 through 10, or actual part numbers).
• Operator - operator identifier (A, B, C).
• Measurement - the recorded measurement value.

Each row is one measurement. Ninety rows total for the standard design.

Step 2: Open the Gage R&R dialog

Navigate to Stat > Quality Tools > Gage Study > Gage R&R Study (Crossed). Use the crossed study when every operator measures every part. Use the nested study only when parts are destroyed during measurement (destructive testing) or when operators cannot share the same parts.

Step 3: Assign the columns

In the dialog:

• Part numbers: select your Part column.
• Operators: select your Operator column.
• Measurement data: select your Measurement column.

Step 4: Choose the analysis method

Select the ANOVA method, not the Xbar and R method. ANOVA partitions variance correctly when there is interaction between parts and operators, which is the realistic case in most manufacturing studies. The Xbar and R method is older and ignores interaction.

Step 5: Set the study variation and process tolerance

Click Options. Two key settings:

• Study variation: leave at 6 (this corresponds to 99.73 percent of normal variation).
• Process tolerance: enter the engineering tolerance (USL minus LSL) if you want percent of tolerance reported. This is critical for go/no-go decisions tied to spec limits.

Click OK twice to run the study.

Reading the Minitab output

Minitab produces a session window summary and six graphs. The numbers that matter most live in two tables: the variance components table and the study variation table.

Variance components and percent contribution

This table breaks total variance into:

• Part-to-part variation.
• Repeatability (equipment variation, same operator, same part, multiple trials).
• Reproducibility (operator variation, different operators, same parts).
• Total Gage R&R (repeatability plus reproducibility).

Percent contribution should add to 100 percent. The widely cited acceptance criterion is:

• Total Gage R&R less than 1 percent contribution: excellent.
• 1 to 9 percent: acceptable.
• Above 9 percent: questionable; investigate before using the gage for SPC or capability work.

Percent study variation and percent of tolerance

The study variation table reports the same components scaled to standard deviation units (multiplied by 6 by default). Look at percent study variation and, if you entered tolerance, percent of tolerance. The common acceptance criteria:

• Less than 10 percent: gage is acceptable.
• 10 to 30 percent: marginal; acceptable for some applications, not for tight capability work.
• Above 30 percent: gage needs work before you can trust capability or SPC results.

Number of distinct categories (NDC)

Minitab reports NDC, sometimes called the discrimination ratio. NDC of 5 or more means the gage can reliably distinguish parts across the range. NDC less than 5 means the gage essentially classifies parts into too few groups to be useful for process control.

The graphs

Six panels appear. The two most useful for diagnosis:

• Components of variation: a bar chart showing how much total variation comes from gage R&R, repeatability, reproducibility, and part-to-part. If the part-to-part bar is small, your parts did not span enough range or your gage is dominating the result.
• Operator by part interaction: lines for each operator across the parts. Parallel lines mean operators are measuring consistently. Crossing lines suggest operator-by-part interaction, which is a real reproducibility problem.

For a deeper look at how measurement noise shows up in capability indices, our breakdown of Cp, Cpk, Pp, and Ppk explains why a marginal Gage R&R can quietly inflate or deflate every capability number you report.

Common mistakes

• Picking parts that all sit near nominal. The gage looks worse because part-to-part variation is artificially small.
• Skipping randomization. Operators remember the previous reading and unconsciously match it.
• Using the Xbar and R method by default. ANOVA is the better choice for most studies.
• Reporting only the percent study variation and ignoring NDC. A study can pass the percent rule and still fail discrimination.
• Treating a Gage R&R as a one-time event. Re-run when fixtures change, after gage calibration, or when a new operator joins the line.

In quality reviews, the most common mistake is treating the percent study variation number as a single pass/fail flag. The graphs and NDC almost always tell a richer story.

Practical action block

What to check before running the study:

• Are the ten parts representative of the full process spread?
• Have operators been briefed and is the measurement procedure documented?
• Is the gage in calibration?
• Is the data collection order randomized?

What to check before accepting the result:

• Did the part-to-part variation dominate the components of variation chart? If not, recheck part selection.
• Are the operator-by-part interaction lines roughly parallel?
• Does NDC reach 5 or more?
• Are you using the right tolerance for percent of tolerance?

When to re-run the study:

• After a gage replacement, repair, or significant calibration change.
• When a new operator joins or the measurement procedure changes.
• When SPC charts show variation that does not match process behavior, suggesting measurement noise.

If your team is using the Gage R&R output to drive root cause work, our structured root cause approach for quality managers covers how to translate measurement findings into action.

Key Takeaways

• Use the crossed ANOVA method in Minitab unless your study is destructive.
• Plan ten parts spanning real process variation, three operators, three trials.
• Look at percent study variation, percent of tolerance, NDC, and the interaction graph together.
• Less than 10 percent study variation is acceptable; 10 to 30 is marginal; above 30 needs work.
• Re-run after gage changes, operator changes, or when SPC behavior suggests measurement noise.