AQL sampling: what the numbers mean

The letters stand for acceptance quality limit. Many buyers, and a great many websites, still expand them as "acceptable quality level", which is the older wording. The change matters because of how the two readings behave in a negotiation: a level sounds like a target to be met, and a limit is the edge of what would be tolerated. Nobody should be aiming at the AQL. It is the boundary at which you stop accepting.

And check which edition anyone is citing. ISO published a THIRD edition of ISO 2859-1 in January 2026, which in ISO's own words "introduces skip-lot sampling procedures, updated guidance for applying sampling strategies, and replaces the previous edition from 1999 along with its amendments". A guide, an inspection company's brochure or a purchase order that cites the 1999 edition is citing a superseded document — which is worth knowing before you write a standard number into a contract.

A sampling plan is built from three things: how big the lot is, how much inspection you want relative to the lot (the inspection level), and the AQL. Those give a sample size and an acceptance number — the maximum number of defective units the sample may contain for the lot to be accepted. That is the whole machine.

This page does not reproduce the tables. They are ISO's copyrighted work, and there is a second reason that matters more here: a table on the screen is what lets a reader skip the sentence that decides everything, which is which part of the standard they are in. Buy the standard, or have your inspection provider state in writing which document, which edition, which inspection level and which AQLs they are applying.

In practice defects are usually split into classes — critical, major, minor — with a different AQL for each and a different consequence attached. No numbers are recommended here: what belongs in each class, and what a failure costs, depends on the product and the market it is going to. That is a decision to make deliberately with your inspector before production, not to copy from a web page.

ISO 2859-1 is titled, in full, "Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot inspection". ISO describes it as "designed for any series of lots produced by a single process or supplier, making it ideal for ongoing production", and says it "includes switching rules that monitor inspection results over time".

Those switching rules are not decoration; they are where a large part of the protection lives. Results are tracked across lots. Consistently good quality earns reduced inspection; deteriorating quality forces tightened inspection, and if it continues, the standard has you stop accepting on sampling altogether. The scheme assumes you will be back, and it punishes a supplier whose quality drifts.

Now apply that to a typical import. One order. One lot. A supplier you have not used before and may not use again. There is no history to switch on and no future lot to be tightened. The sample size comes from a scheme whose teeth are in a sequence that will never exist.

ISO already knows this, and publishes a different part for it. ISO 2859-2 provides plans "for lots in isolation (isolated sequences of lots, an isolated lot, a unique lot or a short series of lots), where switching rules, such as those of ISO 2859-1, are not applicable". Its own list of the situations it exists for reads like a description of small-scale importing: production is intermittent rather than continuous; production comes "from several different sources in varying quantities, i.e. 'job lots'"; lots are isolated; lots are resubmitted after inspection.

Part 2 is indexed differently, and the difference is the useful bit. Instead of an AQL — a limit on the average of a series — its plans are indexed by limiting quality, and ISO states the consumer's risk directly: "the probability of acceptance at the LQ is usually below 0,10 (10 %)". That is a statement about YOUR risk on THIS lot, which is the question a one-off buyer is actually asking.

None of this makes an AQL inspection worthless on a single order — it is far better than no inspection, and the industry runs on it. What it means is that the number should be read more modestly than it usually is, and that if the lot genuinely is a one-off, there is a standard written for your case that you are entitled to ask for by name.

It tells you that a sample was drawn according to an agreed plan, and that the number of defective units found in that sample did not exceed the acceptance number for that plan. That is a real and useful fact. It is not the same fact as "this lot is at or below x % defective".

The gap between those two sentences is the whole of sampling statistics. Every sampling plan carries a probability of accepting a lot that is worse than the index, and a probability of rejecting one that is better. Those risks are the price of not opening every carton, and they are known and quantified in the standard rather than hidden — which is exactly why the standard is worth using and worth citing precisely.

A pass also tells you nothing about the defects the plan was not looking for. An inspection checks against a specification and a defect list agreed in advance. A fault that nobody wrote down is not a defect the inspector can record — it is a conversation you will have later, without leverage. So the defect classification is more important than the AQL number, and it is the part buyers spend the least time on.

And it says nothing about the units not in the sample. If the inspector draws from cartons that are easy to reach, the sample is not random and the plan's statistics do not apply at all. Random selection across the lot, and cartons opened from different pallets and different production dates, is not a detail — it is the assumption everything else rests on.

Finally, a pass is not a certificate of compliance. Whether the product may lawfully be sold in your market is a question about test reports and conformity assessment, not about a sampling inspection, and the two are routinely confused by people who ought to know better.

Agree the defect list before production starts, not before shipment. Write down what counts as critical, what counts as major and what counts as minor, with examples, in the specification the factory works to. An inspection is only as good as the list it is inspecting against, and the list is free to write in week one and impossible to impose in week ten.

Put the plan in the purchase order: the standard and its edition, the inspection level, the AQL for each defect class, and who selects the inspector. Leaving it to "we will do an AQL inspection" is leaving four decisions to be made by whoever is under the most time pressure on the day.

Decide the consequence of a fail in advance, in writing, and this is the clause people forget. What happens on rejection — 100 % sorting at the supplier's cost, rework and re-inspection, a discount, or the order refused? Who pays for the second inspection? How many re-inspections before the order is cancelled? A failed inspection with no agreed consequence turns into a negotiation you conduct while your goods sit in someone else's warehouse.

And book it while there is still time to act on it. An inspection that happens after the container is sealed is a report, not a decision point. The useful moment is when production is substantially complete but the goods have not shipped and the balance has not been paid — which is the same moment the payment article argues for, and for the same reason.

Do not let the factory pick and pay the inspector for an inspection you are relying on. Their own quality department checking their own output is a real and valuable activity, and it is not an independent inspection; the two produce documents that look alike and mean different things.