Measurement system analysis (MSA) step by step
The order of studies that proves a gauge can be trusted: resolution, type 1 study, linearity and bias, Gage R&R and attribute agreement, with acceptance criteria.
Updated 2026-10-24
Every decision about a part or a process relies on a measurement. Measurement system analysis (MSA) proves that the measurement is good enough for that decision. Run the studies in this order: each one builds on the previous one.
1. Resolution
The gauge should resolve at least one tenth of the tolerance, preferably one twentieth (5 %). A gauge that reads in steps of 0.01 mm cannot judge a tolerance of 0.05 mm.
2. Type 1 study: is the gauge itself precise and correct?
One reference part with a known value, measured 25 to 50 times. Gives Cg (precision) and Cgk (precision plus bias). Requirement: Cg and Cgk ≥ 1.33 with 20 % of the tolerance. Type 1 gauge study
3. Linearity and bias: is it correct over the whole range?
Five reference parts across the operating range, each measured about 12 times. The bias should not change with size (slope not significant) and should be close to zero everywhere. Linearity and bias study
4. Gage R&R: does it work with real operators and parts?
Ten production parts covering the process spread, three operators, two or three trials each. Splits the variation into repeatability (equipment), reproducibility (operators) and part-to-part.
| %GRR | Verdict (AIAG) |
|---|---|
| < 10 % | Acceptable |
| 10 – 30 % | May be acceptable depending on the application |
| > 30 % | Not acceptable |
The number of distinct categories (ndc) should be 5 or more. Gage R&R calculator (ANOVA and average and range method).
5. Attribute agreement: for visual and go / no-go inspection
At least 30 parts including borderline ones, two or three appraisers, two or three trials, compared with a known standard. Agreement ≥ 90 % and kappa ≥ 0.75 are good. Attribute agreement analysis
6. Stability over time
Measure the same master part regularly (for example daily) and plot the results on an I-MR chart in the capability tool. Points outside the limits show that the gauge drifts and needs recalibration more often.
Common mistakes
- Choosing ten parts that are almost identical. The part variation is then tiny and %GRR looks terrible.
- Operators who know which part they measure, or see their previous result.
- Using % of study variation for an inspection gauge. For accept / reject decisions % of tolerance is the relevant figure.