AQL and ISO 2859-1: the guide to sampling inspection

You ordered 5,000 desk lamps from a Chinese factory. Production is finished, the balance has not been paid and the container ships in ten days. No one will inspect all 5,000 pieces: an inspector will draw a few hundred at random, and that result will decide whether the goods ship.

That decision almost always rests on AQL and the ISO 2859-1 standard. Here is how to read them and how to write them into a contract, whether you import into the European Union, the United Kingdom or North America.

What is ISO 2859-1 sampling?

Sampling inspection is the middle path between inspecting nothing and inspecting everything. It is used when 100% inspection costs too much, takes too long or destroys the product. As the NIST (National Institute of Standards and Technology, the US federal standards institute) reminds us, its purpose is to decide whether a batch is probably acceptable, not to measure its quality precisely.

"By attributes" means that each unit is classified as conforming or nonconforming: the seam is complete or it is not, the lamp turns on or it does not. This is the inspection mode used in pre-shipment inspections of consumer goods.

The tables descend from the US military standard MIL-STD-105 (last version, 105E, in 1989; canceled in 1995). Its civilian equivalent, ANSI/ASQ Z1.4, has since been the recommended reference in the United States. ANSI stands for the American National Standards Institute, the US standards body, and ASQ for the American Society for Quality; the 2003 edition was reaffirmed in 2018.

Internationally, it is ISO 2859-1. The two standards are very close without being identical (the 1999 ISO revision changed some acceptance and rejection pairs): a contract therefore cites only one standard, with its edition.

What AQL means, and what it does not mean

In the standard, AQL is the worst average production quality tolerated when a supplier delivers a continuous series of batches. It is most often expressed as a percentage of nonconforming units: "AQL 2.5" therefore means, over time, no more than 2.5% defective units on average.

Three misunderstandings come up again and again.

• "I allow 2.5% defects." No. The standard states that setting an AQL does not give the supplier the right to knowingly deliver a single nonconforming unit. A batch at the AQL level does have a good chance of being accepted, but that level is not a target: the system pushes the supplier to do significantly better.
• "The threshold is the AQL multiplied by the sample." Not that either: 2.5% of 200 units is 5, but this plan accepts the batch up to 10, because it is built to almost always accept a batch whose actual quality is at the AQL level.
• "Inspecting 10% of the order is enough." The tables do not work in percentages: the sample size grows far more slowly than the batch. The ISO 2859 series also advises against fixed-percentage sampling, whose risks are unknown.

The three settings of a sampling plan

1. Batch size

This is the number of units presented for inspection. Each product (SKU) in the order forms a separate batch in principle, with its own sample.

2. Inspection level

It sets the sample size. The standard offers three general levels: unless stated otherwise, level II applies; level I inspects fewer units (less precise decision), level III more (safer decision).

Four special levels, S-1 to S-4, give very small samples when a higher risk is accepted: destructive testing, long tests, detailed inspection of packing, loading supervision.

Do not confuse these levels with normal, tightened or reduced inspection, which sets the severity of the plan based on the supplier's history (see below): level III is not "tightened inspection," it simply inspects more units.

Sample sizes for a single sampling plan; the letter in parentheses is the code letter, explained below. For 5,000 units at level II, you inspect 200 units (code letter L), or 4% of the batch; for 50,000 units, 500, or 1%.

3. One AQL per defect class

You set one AQL per class, because a loose thread is not as serious as a bare wire. It is chosen from a standardized series: 0.010; 0.015; …; 1.0; 1.5; 2.5; 4.0; 6.5; 10. When counting nonconforming units, 10 is the maximum; the higher values in the tables are used only to count nonconformities per 100 units.

Critical, major and minor defects

The standard recommends, without requiring it, classifying nonconformities by severity (classes A, B, C…), and leaves it to the parties to define these classes. Most inspections use three.

For demanding products, practitioners tighten these, for example to 0 / 1.5 / 2.5, or even 0 / 1.0 / 1.5. Any deviation from the mandatory safety rules of your market (EU, UK, US, Canada) is classified as critical. For Europe, our guide on CE, UKCA and GPSR product compliance details these requirements.

Reading the tables: code letter, sample size, Ac and Re

Two tables are enough for a single sampling plan under normal inspection, the most common case in pre-shipment inspection. We use their numbering from the 1999 edition and ANSI/ASQ Z1.4; in the 2026 edition, they appear in clauses 15 and 16.

1. Table 1 cross-references batch size and inspection level: it gives a code letter, from A to R (the letters I and O are not used).
2. Table 2-A associates each letter with a sample size and, for each AQL, two numbers: Ac, the maximum number of nonconforming units to accept the batch, and Re, the number at which you reject it (in a single sampling plan, Re = Ac + 1).

Extract from Table 2-A (single sampling plan, normal inspection), read as “Ac / Re.” With code letter L and an AQL of 2.5, the batch is accepted up to 10 nonconforming units and rejected from 11.

Arrows and “AQL 0”

For very low AQLs, some cells contain an arrow. It means: the table does not provide a plan for this combination of letter and AQL, so follow the arrow to the first available plan in the same column. Downward, you apply the plan below with its own sample size (if it reaches the batch size, you inspect everything); upward, the plan above.

Example: code letter L, AQL 0.10 → arrow upward → 125 units, Ac 0 / Re 1. The sample size can therefore vary from one defect class to another.

“AQL 0” does not exist in the tables, whose smallest value is 0.010. It is a contractual convention: Ac 0 / Re 1, a single critical unit found rejects the batch. The standard also provides for a reservation on critical nonconformities: the responsible authority designated by the parties (in practice, most often the buyer) may require every unit to be inspected on these points and reject the batch as soon as the first one is found.

Worked example: 5,000 desk lamps

Run the calculation again with your quantities: the calculator below applies these two tables (1999 edition values), arrows included. Reminder: its levels I, II and III set the sample size, not the severity of the inspection.

The limits of AQL: a statistical tool

Every plan carries two risks: supplier's risk (rejecting a batch that was good) and buyer's risk (accepting a batch that was bad). The operating characteristic curve, or OC curve, gives the probability of accepting a batch based on its actual defect rate. Here is what it gives for the plan in our example (200 units, Ac 10), calculated by us using the binomial distribution.

A batch twice as bad as the AQL (5%) still passes more than half the time. And "AQL 0" does not guarantee the absence of critical defects: out of 200 units inspected, a batch containing 0.5% (25 lamps out of 5,000) has more than a one-in-three chance (about 37%) of being accepted.

The standard is designed for a continuous series of batches, with switching rules (tightened inspection if quality deteriorates, reduced if it stays good, suspension in case of drift) that few buyers apply. For an isolated batch, the 1999 edition advised checking the plan's operating characteristic curve or using ISO 2859-2, whose plans are indexed on a limiting quality. Normal lot-by-lot inspection works, provided you know which risk you are accepting.

Finally, AQL sorts batches, it does not reduce defects. Prevention comes from the factory audit before the order and from inspections before and during production. The points the standard leaves aside (defect list, number of cartons to open, packaging, markings) are settled in the contract: see below.

If the batch is rejected: sorting, rework, counter-inspection

The standard leaves the fate of a rejected batch to the responsible authority: scrap, sorting with or without replacement, rework, re-evaluation or hold. The batch can only be submitted again once all its units have been re-examined and the nonconforming ones removed, replaced or corrected.

1. The inspector issues a report: photos, classified defects, probable cause.
2. The supplier sorts the entire batch for the defects in question and reworks or replaces the nonconforming units.
3. A counter-inspection draws a new random sample; the buyer chooses the inspection (normal or tightened) and the classes to be re-verified.
4. If the result is still bad, the buyer decides: further rework, written waiver (possibly against a credit) or rejection.

Writing AQL criteria into the purchase contract

An AQL plan is only reliably enforceable against the supplier if it appears in the contract or the purchase order accepted before production. Put in writing:

1. The standard and its edition: ISO 2859-1:2026, ISO 2859-1:1999 if your provider has not yet migrated, or ANSI/ASQ Z1.4-2003 (R2018) if it follows the American standard. Only one, the one your inspector actually applies.
2. The plan: single sampling, normal inspection, level II; a special level (S-2, S-3) for long or destructive tests, with their criterion.
3. One AQL per class: for example 0 / 2.5 / 4.0.
4. The classified defect list in an appendix, with photos and tolerances, and the signed reference sample.
5. The counting rules: a unit counted once, in its most severe class; where applicable, how many minors equal one major.
6. The arrangements: batch finished and packed, cartons drawn at random by the inspector, access to all stock, checks outside the count (packaging, markings, barcodes).
7. The consequences of a failure: sorting and rework at the supplier's expense, counter-inspection at its cost, balance paid on an "accepted" report.

No defect list yet? It is the first deliverable to ask your quality control provider for, even before setting the AQLs.

Related guide

In-factory inspections: PPI, DUPRO, PSI and loading check

Pre-production, during production, final, loading: what each inspection checks, when to book it, how to read the AQL report and accept or reject a batch.

12 min read