AQL Sampling PlanAcceptable Quality Limit Sampling Plan

Acceptance Quality Limit (AQL) is a percentage figure, such as 1.0% or 0.65%, that defines the quality level a sampling plan treats as the boundary of routine acceptance. A plan built around AQL 1.0% is designed so that a lot coming from a process genuinely running at around 1% defective has a high probability of passing — not a certainty, a high probability, because the plan is only looking at a sample, not the whole lot. ISO describes ISO 2859-1 as the international standard that "provides sampling procedures and plans for inspection by attributes, specifically using AQL" so that organisations can assess quality from samples rather than full inspection.

The easy misreading is to treat AQL as a promise about the lot you actually received — "we used AQL 1.0, so this lot is at most 1% defective." It isn't that. AQL describes the process average the plan is indexed to; the plan's job is to give a statistically grounded sample size and decision rule for that quality level, with a known (if imperfect) chance of letting a worse lot through and a known chance of rejecting a genuinely good one. A sampling plan is a risk-management tool, not a defect-rate warranty, and nothing in ISO 2859-1 or ANSI/ASQ Z1.4 is written to say otherwise.

Both ISO 2859-1 and ANSI/ASQ Z1.4 work the same way, in two lookup steps. First, the lot size (the number of units in the batch being inspected) and the chosen inspection level are read against a table to produce a sample size code letter — a single letter that stands in for a sample size without yet fixing it to an AQL. Second, that code letter is read against a second table, together with the AQL the organisation has specified for the defect class being checked, to give the actual sample size and the pair of numbers that decide the outcome: the acceptance number (Ac) — the maximum count of nonconforming units still allowed — and the rejection number (Re), the count at which the lot is rejected.

An inspector following the plan draws the sample size at random from the lot, counts the number of defective (nonconforming) units found, and compares that count against Ac and Re: at or below Ac the lot is accepted; at or above Re it is rejected. QualityInspection.org's explainer of the standard summarises the two tables plainly: the first table is used "to get the code letter," and the second "shows the number of samples to pick, based on the code letter" together with the AQL limits expressed both as a percentage and as the acceptance/rejection numbers the inspector actually applies.

The standards define three general inspection levels — I, II and III — plus four special levels (S-1 to S-4) for situations such as destructive or very costly testing, where only a small sample is practical regardless of lot size. Level II, General Inspection, is the default used unless there's a specific reason to use Level I (lower discrimination, smaller sample, used when the cost of letting a bad lot through is low) or Level III (higher discrimination, larger sample, used for new or higher-risk suppliers). Moving between levels changes the code letter — and therefore the sample size — for the same lot size and AQL, which is why the inspection level has to be a documented decision, not something an inspector varies on the day.

ISO 2859-1, Sampling procedures for inspection by attributes — Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot inspection, is the international standard; ANSI/ASQ Z1.4, Sampling Procedures and Tables for Inspection by Attributes, is its US counterpart, described by ANSI's own standards store as "an acceptance sampling system to be used with switching rules on a continuing stream of lots for AQL specified," providing tightened, normal, and reduced plans for attributes inspection. Both trace back to the US military standard MIL-STD-105E, and the two are close enough in structure and terminology that many practitioners use the names near-interchangeably — but they are maintained by different bodies on different revision cycles, so a procedure that cites one should say which one, and which edition.

ISO periodically revises Part 1: the 1999 edition (with later amendments) was in force for a long time, and ISO's own catalogue lists a newer edition — ISO 2859-1:2026 — described by ISO as replacing the 1999 edition and its amendments and introducing skip-lot sampling procedures and updated guidance for applying sampling strategies. ANSI/ASQ Z1.4-2003 has likewise been kept current through reaffirmation (most recently reaffirmed in 2018, per ANSI's preview listing) rather than a full technical rewrite. A quality system that references "ISO 2859-1" or "Z1.4" in a controlled procedure should name the edition in force and update the citation when the standards body publishes a new one — the mechanics described in this article (code letters, Ac/Re, switching) are stable across editions, but specific table values and inspection-level definitions are the kind of detail that can move between revisions.

There is a related standards family worth knowing about even though this article doesn't cover it in depth: ISO 3951 and ANSI/ASQ Z1.9 do the equivalent job for inspection by variables (measuring a continuous characteristic, such as a dimension or a fill weight, rather than classifying each unit pass/fail). Variables plans generally need a smaller sample for equivalent statistical protection, at the cost of requiring a measurable, approximately normal characteristic rather than a simple go/no-go attribute — so the choice between an attributes plan (ISO 2859-1 / Z1.4) and a variables plan (ISO 3951 / Z1.9) is itself a design decision that depends on what's actually being inspected.

A sampling plan is only as good as the defect classification behind it. Good practice — reflected across quality-organisation guidance on AQL rather than fixed by a single clause — is to classify defects before choosing an AQL, typically into critical (a safety, regulatory, or total-loss-of-function risk), major (the unit won't work properly or won't satisfy its intended use), and minor (a cosmetic or low-impact deviation the end user is unlikely to notice or object to), and to apply a tighter AQL to the more serious class. Using one AQL across all defect types regardless of severity — the same number for a label misprint as for a structural crack — defeats the purpose of having a classification scheme in the first place.

It's common across published AQL guidance to see critical defects held to a very tight AQL (sometimes effectively zero tolerance, with critical defects pulled out of the attribute-sampling framework altogether and handled by 100% inspection, automated detection, or a capability-based approach instead), major defects sampled at a noticeably tighter AQL than minor defects, and minor, cosmetic-only defects given the most permissive AQL of the three. The actual numbers an organisation chooses are a risk decision tied to the product, its use, and its regulatory context — the standards supply the sampling mechanism for a chosen AQL; they don't prescribe which AQL a given defect class must use.

Both standards are built to run as a continuing system across a stream of lots from the same supplier, not as a one-off calculation per lot — ANSI's own description of Z1.4 calls out that it is "an acceptance sampling system to be used with switching rules on a continuing stream of lots," providing tightened, normal, and reduced plans. Normal inspection is the default starting point. When recent lots start failing at a rate that suggests the supplier's process has drifted, the rules call for switching to tightened inspection, which applies a stricter acceptance number for the same sample size (or an otherwise less permissive plan) until the supplier demonstrates it is back in control, at which point the rules provide for switching back to normal.

Reduced inspection works in the other direction: after a run of consistently accepted lots under normal inspection, with the underlying process shown to be stable, the rules allow a smaller sample size for the same AQL as a form of earned credit for demonstrated quality — and a single failure, or a result that doesn't meet the reduced plan's own qualifying criteria, sends inspection back to normal. ASQ's own "Ask the Standards Experts" column on Z1.4 makes the point that moving to reduced inspection isn't just about counting consecutive accepts: the standard's reduced-inspection table carries its own footnote conditions, and a production process being judged "stable" for this purpose is about the process showing control over time, not simply about a capability index crossing some number. The newer edition of ISO 2859-1 (per ISO's own 2026 description) also adds skip-lot sampling procedures, which go a step further than reduced inspection by allowing some lots in a qualifying stream to be accepted without a full fresh sample draw at all, subject to the qualifying conditions the standard sets.

Every sampling plan has an operating characteristic: a relationship between the true (unknown) defect rate of a lot and the probability that the plan accepts it. At the AQL, that probability is high by design; as the true defect rate rises well above the AQL, the probability of acceptance falls, but it is never zero for any finite sample short of inspecting the whole lot — a plan can still accept a bad lot, and it can still reject a genuinely good one, by chance. Larger samples sharpen that relationship (better separation between "probably fine" and "probably not"); they don't eliminate the underlying statistical uncertainty.

That's the core reason records matter as much as the Ac/Re decision itself. A documented plan should capture which AQL and inspection level applied, the lot size, the code letter and sample size actually used, the switching state at the time (normal, tightened, or reduced), the defects found, and the accept/reject outcome — so that if a lot is challenged later, or a customer or auditor asks why a particular result was accepted, the answer is the plan that was actually applied, not a reconstruction after the fact.

In a manufacturing operation, AQL sampling plans most often sit at goods-in: incoming components, packaging, or raw materials are received against a purchase order, and a sampling plan decides how many units from that delivery get inspected and what inspection result releases the material to stock versus holds it for disposition. The lot size used in the lookup is the delivery (or the inspection lot defined by the quality procedure, which isn't always identical to the delivery quantity), and the record of that inspection — code letter, sample size, Ac/Re, defects found, and the release decision — becomes part of the traceable history for every batch that later consumes that incoming material.

That connection matters at batch release, because a finished-batch release decision can only be as sound as the incoming-material decisions underneath it. If a goods-in lot was accepted under a tightened plan because the supplier's recent performance had already slipped, a reviewer assessing whether a finished batch is fit to release benefits from seeing that context — not just a flat "accepted" stamp with no indication of which plan or switching state produced it. The same logic applies to in-process and finished-goods sampling further down the line, where AQL plans are sometimes used for packaging-line checks or outgoing inspection before a batch record is closed out.

• Treating AQL as a target or a warranty — reporting "this lot met AQL 1.0" as though it proves the lot is better than 1% defective, rather than describing the sampling risk that was accepted.
• Applying one AQL across critical, major, and minor defects instead of classifying defects first and setting a tighter AQL for the more serious classes.
• Running every lot and every supplier on normal inspection indefinitely, never switching to tightened when performance slips or earning reduced/skip-lot inspection through demonstrated stability.
• Citing "ISO 2859-1" or "Z1.4" in a procedure without naming the edition, so the controlled document can drift out of step with the current published standard.
• Losing the link between the sampling record (lot size, code letter, sample size, Ac/Re, switching state, result) and the batches or shipments that consumed the inspected material, which makes it hard to reconstruct the basis for a release decision later.
• Using an attributes plan (ISO 2859-1 / Z1.4) by default when a measurable characteristic and a variables plan (ISO 3951 / Z1.9) would give equivalent protection from a smaller, cheaper sample.