AQL sampling in pre-shipment inspection: how acceptance sampling actually works (ISO 2859-1)

How the mechanics work: from lot size to accept/reject numbers

ISO 2859-1 is the international acceptance-sampling standard descended from MIL-STD-105E. It is published in the United States as ANSI/ASQ Z1.4 and adopted in China as GB/T 2828.1 — which matters in practice: the QC office of any organised Chinese factory knows the same tables under the GB designation, so citing either reference puts both sides on identical numbers.

The procedure has five steps. Worked example: a purchase order for 400 DN50 stainless steel ball valves, presented for pre-shipment inspection.

1. Define the lot. The lot is the population the result applies to — normally units of the same model, rating and drawing revision under one purchase order line, produced under uniform conditions and presented for inspection at one time and place. Do not let unlike items be combined into one inspection population to make the numbers easier.

2. Choose the inspection level. General Inspection Level II is the default for pre-shipment inspection in China. Level I draws a smaller sample, Level III a larger one, and special levels S-1 to S-4 give very small samples for checks that are slow, costly or destructive — a test that consumes a sealed assembly, for instance.

3. Find the sample size code letter. The standard’s first table crosses lot size with inspection level. A lot of 281–500 units at Level II gives code letter H.

4. Read the sample size. Code H means 50 units, which the inspector must select at random across cartons and serial numbers — not from whichever boxes the factory puts forward. If the factory lines up fifty “good” units in a clean area before the inspector arrives, that is not AQL sampling.

5. Read the accept/reject numbers. In the single-sampling, normal-inspection master table, the row for code H crossed with each AQL gives the acceptance number (Ac) and rejection number (Re). At AQL 2.5: accept on 3, reject on 4. At AQL 4.0: accept on 5, reject on 6. Re is always Ac + 1 — there is no grey zone in single-sampling normal inspection. Each defect class is judged separately; if any class reaches its rejection number, the whole lot fails. Common commercial practice counts a defective unit once, in its worst defect class.

The plans are built so a process running exactly at the AQL passes most of the time — the producer’s risk is kept small; the mirror image, the consumer’s risk, is covered in the FAQ. Switching rules (normal to tightened after two rejections in five lots, reduced after a sustained clean run) also exist, but presuppose a continuing series of lots — relevant on repeat orders, largely irrelevant on a one-off industrial purchase order.

The AQL chart for the worked example

For the 400-valve lot, General Level II, single sampling, normal inspection:

Plain-English read: out of 50 units, the lot of 400 passes if the inspector finds zero critical defects, three or fewer majors, and five or fewer minors. Four majors, six minors, or a single critical, and the lot fails. Note that 3 majors in 50 is 6 per cent — the table is not saying 6 per cent is acceptable; it is applying a statistical lot-acceptance rule indexed to AQL 2.5. That is why “AQL 2.5” is so widely misread.

And the surrounding lot-size bands, for the two customary AQLs:

Values follow the ISO 2859-1 single-sampling table for normal inspection; confirm any cell against the current edition of the standard before writing it into a contract.

Critical, major and minor defects — who sets the three numbers

Sampling runs per defect class, and the classes mean roughly this:

• Critical — creates a safety hazard, breaches a statutory or certification requirement, or renders the product unusable. On industrial goods: missing earth bonding, wrong voltage configuration, a pressure part without its required certificate, an unapproved material substitution.
• Major — impairs function, durability or acceptability; likely to cause failure in service or rejection by the customer. On industrial goods: wrong gasket material, an IP rating below specification, coating below minimum dry-film thickness, incorrect tagging against approved drawings.
• Minor — a workmanship or appearance deviation that does not affect use. A small paint blemish in a non-critical area is minor. A document error that blocks site acceptance is not minor just because it is “paperwork”.

The standard does not assign AQL values to these classes. The buyer does — and where the buyer is silent, the inspection company’s default applies, customarily 0 / 2.5 / 4.0 for critical / major / minor on general merchandise.

Industrial buyers routinely tighten this. “Critical: zero” does not mean an AQL of 0 — no such column exists — but a contractual rule: any critical defect in the sample fails the lot outright and triggers 100 per cent screening of the affected feature. The logic is consequence, not statistics: on industrial equipment a critical is a pressure boundary, an earthing continuity, a rating mismatch, and the cost of one escape is out of all proportion to the unit price. Many buyers on project critical paths also pull majors from 2.5 down to 1.5 or 1.0 — and, more importantly, define defects against the technical specification rather than against retail cosmetic criteria.

Where AQL breaks down for industrial equipment

The tables assume a large population of interchangeable units flowing from a homogeneous process. Industrial procurement violates each assumption in turn.

Small lots. Two compressors, four transformers, six pressure vessels: lot sizes of 2 to 8 give code letters A or B and samples of 2 or 3, which discriminate almost nothing. The standard closes the loop itself — where the sample size meets the lot, you inspect 100 per cent.

Serialised units. Each machine carries its own nameplate data, mill certificates and factory test record. Conformity is established per serial number; sampling a population of four individuals is a category error.

Value asymmetry. Sampling is an economic trade-off between inspection cost and the cost of escaped defects. When one unit is a six-figure machine whose failure stalls a commissioning schedule, the trade inverts and 100 per cent inspection is cheap.

No series. AQL describes a tolerable process average across continuing lots. A one-off purchase order has no series, no switching rules, no average.

Function, not attributes. AQL counts defects an inspector can see, measure or quickly check on a sampled unit. Flow, pressure, torque, accuracy and control sequences have to be demonstrated under a defined test protocol — that is a type-test or witnessed factory acceptance test problem, not an AQL problem.

What replaces sampling where it does not apply:

The common real-world case is hybrid. One shipment might combine three machines with forty cartons of fittings, gaskets, fasteners and commissioning spares. The machines get per-serial verification linked to the FAT record; the cartons get ISO 2859-1 sampling. A competent inspection scope — and your pre-shipment inspection checklist — should state which regime applies to which purchase order line. The right question is never “can we use AQL?” but “which characteristics can safely be sampled, and which must be witnessed, tested or checked 100 per cent?”

Writing sampling into the purchase order so the inspection is enforceable

An inspection clause that just says “AQL 2.5” is not enforceable. When a lot fails, you do not want to be arguing the definition of “major” with the supplier’s quality manager while the vessel is booked. Seven elements make the difference:

1. Standard and plan, by name and edition. “Acceptance sampling per ISO 2859-1 (GB/T 2828.1), single sampling plan, normal inspection, General Inspection Level II.” Name the edition you mean; do not leave the inspector and supplier to guess which table applies.

2. Lot definition. Units of the same model, rating, drawing revision and purchase order line, completed and offered for the same shipment release — production 100 per cent complete and, by common practice, at least 80 per cent packed before the inspection is booked.

3. AQLs per defect class, as numbers. Critical — none permitted; major — AQL 2.5 (or 1.5/1.0 on critical-path equipment); minor — AQL 4.0. “Standard AQL applies” is not a specification.

4. A defect classification annex. A product-specific table mapping concrete defects to the three classes. Without it, the inspector’s generic checklist decides for you.

5. Counting rule. State whether you count defective units or individual defects. Commonly, one unit with several major defects is one major defective unit for the arithmetic — but the report should list every defect found.

6. Sampling authority. The inspector selects units at random across cartons and serial numbers — the square-root-of-cartons-plus-one drawing rule is common practice — with unrestricted access to the full lot. Factory-preselected or sealed samples void the result.

7. The override clause. This is the one that prevents the classic supplier argument:

And state the failure terms: a failed lot is held; sorting and rework are at the supplier’s cost, as is re-inspection (optionally under tightened acceptance numbers); and release — balance payment or shipping authorisation — is conditioned on a passed inspection with written buyer authorisation.

Frequently asked questions