Academy › Practical guides › Quality tools
The 8 control-chart out-of-control rules — and when enabling all of them is a mistake
Nelson / Western Electric patterns, what each one detects, and the false-alarm cost nobody mentions.
Updated 13 November 2026 · 7 min read · Written for working labs and quality teams
Short answer
The eight commonly used run rules flag non-random patterns beyond a single point outside 3σ: 1 point beyond 3σ; 9 points on one side of the centre line; 6 steadily increasing or decreasing; 14 alternating up and down; 2 of 3 beyond 2σ; 4 of 5 beyond 1σ; 15 points within 1σ of the centre line; and 8 points with none within 1σ. Each detects a different assignable cause — but enabling all eight on every chart raises false alarms sharply, so choose the rule set per process and document it.
On this page
The rule set at a glance
These are the widely used Nelson rules, descended from the Western Electric handbook. Numbering and exact counts vary slightly between software packages and AIAG/SPC references — write down the rule set you actually use in the control-plan or procedure, and keep it consistent across shifts. The pattern logic matters more than the numbering.
| Rule | Pattern (one side of centre line unless stated) | Usual assignable cause |
|---|---|---|
| 1 | 1 point beyond 3σ | A sudden special cause, or a data-entry error |
| 2 | 9 consecutive points on the same side | A sustained mean shift |
| 3 | 6 consecutive points steadily increasing or decreasing | Tool wear, drift, warm-up, depletion |
| 4 | 14 consecutive points alternating up and down | Systematic over-adjustment / two alternating sources |
| 5 | 2 of 3 consecutive points beyond 2σ (same side) | A moderate shift starting |
| 6 | 4 of 5 consecutive points beyond 1σ (same side) | A smaller shift starting |
| 7 | 15 consecutive points within 1σ of the centre line | Stratification: rounded data, or mixed streams pooling |
| 8 | 8 consecutive points with none within 1σ (both sides) | Mixture: two machines, cavities or suppliers pooled |
Reading rules 7 and 8 correctly
Rules 7 and 8 are counter-intuitive and are where practical experience shows up. Rule 7 looks like "wonderfully tight control", but fifteen points hugging the centre line usually means the data are rounded too coarsely, the gage resolution is inadequate, or someone is averaging across streams before plotting. Rule 8 looks like "wild variation", yet it is often two perfectly stable streams — two cavities, two spindles, two suppliers — plotted on one chart. Splitting the chart by stream resolves it; tightening the process does not.
Rule 4 (alternating) is the signature of operator over-adjustment: every time the chart moves, someone nudges the setpoint. The fix is to stop adjusting to common-cause noise, not to narrow the control limits.
Choosing which rules to enable
- Rule 1 alone has a false-alarm probability of about 0.27% per point for a stable normal process. Adding rules 2, 5 and 6 increases sensitivity to small shifts — and also increases the total false-alarm rate. More sensitivity is not free.
- High-volume, low-consequence processes: rule 1 plus rule 2 is often enough; add 5/6 when a small shift is expensive to miss.
- Safety or regulated characteristics: enable the documented rule set the customer requires, and make sure the reaction plan says what to do for each signal.
- Always pair a rule with a reaction plan. A signal with no defined response — contain, check the last good point, investigate, record — is just a red dot on a screen.
Practical note. Control limits are calculated from process data; specification limits come from the customer and describe what the product must do. Never draw a spec line on an X̄ chart as a control limit, and never recalculate limits from a period you already know contained a special cause.
Before the first point goes on the chart
- Rational subgrouping. Sample so that within-subgroup variation represents common-cause noise only. Five consecutive parts from one setup is a subgroup; five parts spread across a shift is not.
- Measurement system check. A poor gage R&R inflates the limits and hides real signals, or creates apparent ones.
- Phase I vs Phase II. Use an initial baseline run to estimate limits with known special causes removed; then freeze those limits for monitoring. Quietly re-fitting limits every week makes the chart unable to detect drift.
- Chart type matches the data. X̄-R / X̄-s for subgrouped variable data, I-MR for individual measurements, p / np for pass-fail proportions, c / u for counts. Capability analysis comes after stability, not before.
Where this goes wrong in practice
Enabling all eight rules on every chart and then ignoring the alerts. Alarm fatigue is worse than a slightly less sensitive chart.
Using specification limits as control limits, which makes an incapable process look "in control" and vice versa.
Pooling two machines or cavities into one chart and then investigating rule 8 as if it were instability.
Recalculating control limits after every batch, which slowly bakes drift into the baseline.
Investigating a rule-1 signal without first checking whether the value was transcribed correctly — a surprising share of "special causes" are transcription errors.
Frequently asked
Are these rules required by ISO 9001 or IATF 16949?
No. The standards expect you to monitor and control processes and to react to special causes; the specific run-rule set is normally defined by the customer, the AIAG SPC reference manual, or your own control plan. Document which set you use.
What is the difference between a control limit and a specification limit?
A control limit describes what the process is actually doing (calculated from process data), while a specification limit describes what the customer requires. A process can be stable and incapable, or capable on average and unstable — the two answers come from different charts.
How many points do I need before control limits are trustworthy?
Common practice is roughly 20–25 subgroups (about 100 individual values) for a Phase I baseline, with known special causes investigated and removed before the limits are frozen. Fewer than that gives wide, unstable estimates.
Can I use run rules on an I-MR chart with individual measurements?
Yes, most software applies rules 1–4 and 6 to individuals charts, but rules involving subgroup averages have different behaviour and autocorrelation in continuous-process data can generate false signals. Check the rule definitions in your software before applying them.
Plot it, don't guess it
Build X̄-R, I-MR and p charts with live out-of-control flags
Paste your sample data, pick the chart type and rule set, and see the signals flagged as you enter values — including which rule fired, so you can write a reaction plan against it.
Open the free SPC control-chart tool →Runs in your browser. Export the report free. No sign-up needed.
LabKitPro guides are written from published standards and accreditation-body documents — references, clause numbers and paraphrased intent only, never verbatim standard text. They are for preparation and study, not a certification or accreditation decision, and not legal advice. ISO standards are trademarks of ISO; IATF 16949 is published by the International Automotive Task Force and is not an ISO standard. The current laboratory standard is ISO/IEC 17025:2017; there is no 2025 edition. Global ACI replaced IAF and ILAC on 1 January 2026.
Academy by standard · All practical guides · Free browser tools