ISO 7870 — the control chart standard, part by part

The international framework for control charts, prepared by ISO/TC 69/SC 4. What each of the six parts covers, which one you actually need, and which of them this platform implements.

ISO · Our reading of it — the source document is the authority. The document itself ↗

Last reviewed 2026-08-18

If you want the neutral, international, vendor-free definition of what a control chart is and how it should be used, it is the ISO 7870 series, prepared by ISO technical committee TC 69 (Applications of statistical methods), subcommittee SC 4 (Applications of statistical methods in process management).

Almost every sector-specific document — the AIAG & VDA SPC manual, the healthcare guidance, the pharmaceutical expectations — either aligns with ISO 7870 or quietly restates it. It is the backbone. It is also, like all ISO standards, a paid document: what follows is our reading of what each part is for, so you can buy the right one instead of all six.

The six parts

Part Covers Current edition
7870-1 General guidelines. The principles, the vocabulary, and how the various chart approaches relate — the part that helps you choose the right one. 2019
7870-2 Shewhart control charts. The classic X̄-R, X̄-s, I-MR and attribute charts, plus warning limits, trend/pattern tests and process capability. 2023
7870-3 Acceptance control charts. For processes deliberately allowed to move within a tolerance: sample sizes, action limits, decision criteria. 2020
7870-4 CUSUM charts. Cumulative sum schemes for variables and attributes — the tool for detecting small, sustained shifts quickly. 2021
7870-5 Specialized control charts. For when the Shewhart approach is inapplicable or inefficient — autocorrelated data, short runs, mixed processes. A revision is in draft. 2014
7870-6 EWMA charts. Exponentially weighted moving average, for small shifts in the mean of a continuous process. 2016

For most people the answer is part 2 and nothing else. Buy part 1 as well if you are writing a company SPC procedure and need the selection logic to cite. Parts 4 and 6 matter when Shewhart charts are demonstrably too slow for the shift size you care about; part 5 matters when your data breaks the assumptions rather than the process.

Part 2 is the one that describes what most people mean by "SPC"

The Shewhart part is where the familiar machinery lives:

  • Limits at three sigma from the centre line, estimated from within-subgroup variation — R̄/d₂ for a range chart, s̄/c₄ for a sigma chart, MR̄/d₂ for individuals — not from the standard deviation of all the data. This distinction is the whole reason a control chart can detect a shift it has already absorbed into its own arithmetic if you get it wrong.
  • The chart constants (A₂, D₃, D₄, d₂, A₃, B₃, B₄, c₄) as functions of subgroup size. We publish them as a free lookup table with the formulas.
  • Pattern tests — runs, trends and the rest — as a supplement to the three-sigma rule, with the explicit warning that applying many of them at once raises the false-alarm rate.
  • Process capability treated as a separate question from process control, which is the point everyone in a factory argues about at least once a year.

What ISO 7870 does not do

It does not certify anything, and there is no such thing as being "ISO 7870 certified". It is a methods standard: your quality system may reference it, an auditor may ask whether your charts follow it, but nobody audits you against it directly. Anyone selling you "ISO 7870 compliant software" is describing a chart drawn correctly, which is a floor rather than a feature.

It also does not tell you what to chart. Choosing the characteristics is yours, and it is the decision that determines whether any of this is worth doing.

What this platform implements

  • Part 2, Shewhart charts — X̄-R, X̄-s, I-MR and c-charts, with limits from within-subgroup variation using the standard constants. The engine's test suite reproduces published worked examples, so the numbers are checkable rather than asserted.
  • Pattern tests as the eight Nelson rules, switchable per characteristic with a severity each — so you can follow part 2's warning about false alarms by turning on only the rules you want to act on.
  • Capability — Cp, Cpk, Pp and Ppk, reported side by side with the sigma each used, because quoting one without the other is how capability numbers mislead.

And what it does not:

  • No CUSUM (part 4) or EWMA (part 6). If your process needs to detect a half-sigma shift quickly, these are the right charts and we do not draw them yet.
  • No acceptance control charts (part 3).
  • No specialized charts (part 5) — no short-run/standardised charts, no autocorrelation handling, no multivariate methods.
  • No p, np or u charts. Attribute data is limited to counts on a c-chart.

If you would like to see the maths before believing any of it, the free tools run the same engine as the product, with no signup and nothing stored.

Who this applies to

  • SPC for automotive manufacturing — What IATF 16949 and the new AIAG & VDA SPC manual actually ask of a production line, why the capability number on the PPAP is not the point, and where a live chart earns its keep.
  • SPC for pharmaceutical manufacturing — Continued process verification is SPC with a regulator attached. What FDA Stage 3 and EU GMP Annex 15 expect on a chart, and what a small tool can and cannot do inside a GMP quality system.
  • SPC for semiconductor manufacturing — Fabs invented most of modern SPC practice and generate more process data than anyone. What is charted, why the range chart is often the interesting one, and where a small tool is and is not the right answer.
  • SPC for food and beverage manufacturing — SPC inside HACCP monitoring, net-content control that satisfies the weights regulations without giving product away, and where a live chart beats a paper checksheet.
  • SPC for aerospace manufacturing — AS9103 asks for the variation of key characteristics to be managed, not just inspected. What that means on a low-volume line, and how to run control charts when a "batch" is eleven parts.

Charts that hold still long enough to be audited

Limits computed from a baseline window and frozen there, every change to them recorded with who made it and when, the rule set stated on the chart, and a read-only view anyone can be sent. That is most of what these documents ask a tool to do.