AQL Sampling for China Inspections: How to Set Limits That Actually Protect Your Order (2026)

Almost every inspection booking form in China arrives pre-filled with "AQL 2.5 / 4.0." It is the industry default, it is rarely chosen deliberately, and for a meaningful share of products it is the wrong answer.

I have sat through this hundreds of times. A buyer books an inspection because they are nervous, the company sends a form, the buyer ticks the box that is already ticked, and three weeks later a green PASS stamp arrives. The container ships. Then the returns start, or the chargeback arrives, or the reviews mention a button that sticks.

The report was not wrong. The buyer never decided what "passing" meant.

1. What AQL actually measures

AQL stands for Acceptance Quality Limit. The formal definition, in the ANSI/ASQ Z1.4 and ISO 2859-1 family of standards, is the worst tolerable process average when a long run of lots is submitted for acceptance. In plain terms: it is the defect rate at which a lot still has about a 95% chance of being accepted.

Notice what that sentence does not say. It does not say "the maximum defect rate you will receive." A lot sitting exactly at AQL passes nineteen times out of twenty. A lot running at twice AQL still passes often. Sampling is a filter with deliberate holes in it, because 100% inspection of every unit is slow, expensive and — on cosmetic defects — unreliable anyway.

So the first decision is not the number, but the question: what is a defect? Until that is written down, 2.5 means nothing, and the inspector classifies by their own judgment — not yours, and not your customer's.

2. Lot size sets the sample, not your budget

Sample size is a function of lot quantity and inspection level. Most firms default to General Inspection Level II, which produces this shape. The full standard has arrows that redirect some combinations; confirm exact values with the company doing the inspection.

Two things surprise new importers.

First, the sample barely grows with the order. A 2,000-unit order and a 9,000-unit order are inspected at 125 and 200 pieces respectively. That is the point of acceptance sampling: you are buying a statistical read on the lot, not a warranty on every carton. If you need unit-level certainty, sampling is the wrong tool and you should be talking about process control instead.

Second, small orders swing harder. At 32 pieces sampled, two major defects pass and three fail the lot. That is fine — but it explains why a boutique 200-unit order flips between PASS and FAIL for reasons that feel arbitrary.

3. Classify defects before you set limits

Every inspection report sorts findings into severity classes. The classes are yours to define, and defining them is the single highest-leverage hour you will spend on quality. If you leave it to the inspection company's default template, you will get their definitions.

Write the tolerances into the specification, not into the conversation. "No visible scratches" is unenforceable. "No scratch longer than 1mm on any surface visible in the unboxed orientation; scratches on the base acceptable up to 5mm" is enforceable, and an inspector can apply it without guessing.

Photographs do more work than adjectives. A one-page defect sheet — good unit, major, minor, each shot on your actual product — kills most disputes before they start. Have the factory sign it with the golden sample.

4. Choosing limits by channel, not by habit

The right AQL depends on what happens to a defective unit after it reaches the customer.

• Amazon FBA and other marketplace sales: a defective unit costs the refund, the return freight, the review damage and potentially the listing's health score. Tighten major defects to 1.5 and treat packaging integrity as major, because a crushed box on a marketplace photo-driven listing is a return even when the product is perfect.
• Retail and wholesale: your buyer's receiving audit is stricter than yours, and chargebacks arrive weeks later. Match your limits to their vendor manual, not to a generic default — if their manual specifies sampling, use their numbers, since your inspection is only a pre-check for theirs.
• Promotional and gift channels: cosmetic consistency matters more than long-term durability. Minor defects can often sit at 4.0 while packaging and print accuracy get promoted to major.
• Regulated categories — children's products, electronics, anything with a battery or a mains connection: critical is zero and stays zero. Compliance labelling, language requirements and safety markings belong in the critical class even though they are "just print."

One warning: AQL 1.0/1.5 on a hand-finished product will fail lots that are commercially fine, and you will spend a month arguing about who pays for rework. Limits that fail good lots cost as much as limits that pass bad ones.

5. Five inspection stages — pick the one that catches your failure

Most buyers book one inspection and expect it to cover everything. Each stage answers a different question.

Pre-shipment inspection is the default because it is the cheapest useful checkpoint. It is also the weakest for a new product: by the time it runs, thousands of units exist. If the defect is in the design or the spec, the inspector will find it faithfully and reject a lot that should never have been started. That distinction — conformance versus suitability — is the one most buyers miss, and it is why the quality control sequence puts the audit and the golden sample ahead of the container.

6. What sampling cannot catch

An AQL inspection is a visual and dimensional read on a sample, plus some functional checks. It is blind to a specific list of failures that happen to be the expensive ones:

• Reliability over time. A unit that dies in week six passes at the factory. Only design review, component verification and life testing address this.
• Hidden internals. Solder quality, wire gauge, adhesive coverage and battery cell provenance are invisible to a cosmetic inspection. Component-level verification and teardown sampling are separate exercises.
• Material substitution. If the factory swaps a resin or a chip, the sample looks identical and behaves differently. The only defence is a locked BOM plus periodic lab testing.
• Compliance. An inspector confirms a label exists and reads correctly. They do not confirm the product passes the standard the label claims.
• Anything the spec never defined. The inspector measures against your document. An empty document yields a report about nothing.

Where you cannot inspect, you audit. How incoming components are checked, how the line is set up, what the factory's own QC does between stations — that predicts reliability far better than a bigger sample. It is also why the sample you approve carries more weight than any report: it is the only artefact that survives the whole run.

7. Reading a report before you accept "PASS"

The stamp on page one is a summary written by a human being working for a company you hired for one day. Read the body. Eleven things to check:

8. How a bad lot gets a good report

None of these need a dishonest inspector. Most need only a buyer who specified nothing.

• The pre-sorted lot. The factory knows the inspection date. Cartons presented for sampling are not necessarily cartons drawn at random. Sampling across pallets and asking for cartons from different production dates is the counter.
• Definition drift. Under a vague spec, "minor cosmetic issue" absorbs defects you would call major. Your defect sheet is the counter.
• Rework without disclosure. Sorted, touched-up units returned to the lot are legitimate only if the rework is disclosed and re-verified. Ask directly: was this lot sorted, and what was pulled?
• The sample that came from somewhere else. Approved samples have a way of being better than production. Sealing a signed golden sample and keeping a duplicate at the inspection company is cheap insurance.

9. The clauses to settle before the first inspection

An inspection only creates leverage if the purchase order gives it teeth. Four terms do most of the work:

1. Right to inspect, at any stage. The PO should state that you or a nominated third party may inspect goods and records before shipment, and that inspection does not constitute acceptance.
2. Defect definitions as an annex. Reference the defect reference sheet and the tolerance spec as binding annexes, not as emails.
3. Failure consequences. Specify what happens on a failed lot: sorting at whose cost, rework deadline, re-inspection fee borne by the factory when the failure is theirs, and your right to reject and recover the deposit.
4. Payment tied to outcome. Balance against a passed inspection or a bill of lading, never against a promised ship date. This is the clause that changes behaviour, because it is the only one the factory's accountant reads.

If a lot does fail, the recovery conversation has its own playbook — evidence, cost quantification, and a remedy ladder from rework to credit to replacement — covered in how to get refunds and compensation from Chinese suppliers. What matters at the inspection stage is that the report is specific enough to be evidence: quantities, photographs, measurements and the standard applied.

10. Cost, and what you are actually buying

Third-party inspection is priced per man-day, plus travel for out-of-the-way factories. Local firms quote lower than the international names; the difference buys brand recognition and more consistent report formats, not necessarily a better inspector. A single pre-shipment inspection on a typical consumer-goods order usually lands in the low hundreds of US dollars, and a re-inspection costs about the same again.

Put that against the alternative. A failed lot discovered at the port costs you freight both ways, storage, the retail window, and the margin on the whole order. The inspection fee is the cheapest line item in the entire transaction, which is why the argument for raising the sample or adding a stage should never be about the fee — but about which failure you are trying to catch.

The Bottom Line

AQL 2.5/4.0 is a default, not a decision. The number is the last thing to choose: define critical, major and minor against your actual product and channel, photograph the tolerances, put the definitions in the PO as an annex, then pick limits that reflect what a defective unit costs you downstream.

And remember what the report is. It is a sample-based read on a finished lot, produced by someone working one day for you. It is strong evidence about conformance and weak evidence about suitability. The design, the spec, the golden sample and the audit are where quality is actually decided — inspection is where you find out whether the factory followed them.

— Jack Zhang, ChinaBizGuides