AQL sampling table: how many units to inspect on your first order

AQL 2.5 does not mean "2.5% of your order can be defective." That reading is wrong in a way that changes what you demand from a factory, because a sample of 80 units with 5 defective ones — 6.25% of the sample — can still pass an AQL 2.5 lot. The useful version of AQL is a lookup: for a 1,200-unit order, the standard says inspect 80 units, accept on 5 or fewer major defects, reject at 6. Everything else — levels, defect classes, what to write in the PO — hangs off that lookup.

AQL — Acceptable Quality Limit — is the worst defect rate a sampling plan is designed to pass routinely. This article explains the lookup, shows the numbers at merchant-sized order quantities, and walks a worked plan for a first order. How the inspection itself runs day-to-day — booking an inspector, the unit-by-unit checks, the report format — is a separate discipline covered in the pre-shipment QC checklist.

What AQL 2.5 means in units at your quantity

Every AQL lookup turns on two numbers. The sample size is how many units get pulled from the lot and examined; the accept number is how many defects that sample may contain and still pass — one more, and the whole lot fails. Here's the table merchants actually need — General Inspection Level II, the default, under normal-tightness single sampling, with the industry-standard settings of majors at AQL 2.5, minors at AQL 4.0, and zero tolerance on criticals (ANSI/ASQ Z1.4, accessed November 2026):

Read one row aloud and the counterintuitive part stops being confusing. An order of 1,000 phone stands: pull 80 units at random from across the finished cartons. If the sample shows 5 majors, the lot passes; the 6th major fails it. Minors run on their own counter: 7 passes, 8 fails. The percentages on the sample are deliberately looser than the AQL number because the plan is calibrated against process averages — a factory running at a true 2.5% major rate passes most lots, while a factory drifting toward 6% fails most of them. The table is a screen for process drift, not a permission slip on individual units (ANSI/ASQ Z1.4, accessed November 2026). One scaffolding note before the worked plan: the columns run on separate counters for the three defect classes — criticals (safety or regulatory failures, where any single one fails the lot), majors (the unit can't be sold or won't survive normal use), and minors (sellable, but a customer might notice). Defining the classes for your product and writing them into the PO is the pre-shipment QC checklist guide's territory; for reading the table, the three counters above are all you need.

Choosing the inspection level

The tables above assume Level II, the default for a reason: it balances inspection cost against the odds of catching a real drift. The other two levels exist for specific situations:

• Level I. A smaller sample — on a 501–1,200-unit lot, 32 units instead of 80. Defensible for repeat orders from a factory with a clean history on a low-value SKU. The trade is blunt: fewer eyes, smaller chance of catching moderate drift.
• Level III. A bigger sample — 125 on the same lot. The choice for a first order with a new factory, a high-value product, or anything where a hidden defect rate is expensive to discover after receipt.

The standard also defines switching rules — tightened sampling after failures, reduced sampling after sustained passes — but a merchant running a handful of orders a year can ignore the machinery and just choose the level per order deliberately (ANSI/ASQ Z1.4, accessed November 2026). The decision in one line: history and the price of discovery decide. A factory with a clean track record on a low-value reorder earns Level I; a first order from an unproven factory, a high-value product, or anything whose hidden defect rate is expensive to find after receipt takes Level III; everything in between stays at Level II. When in doubt, the cost asymmetry settles it — the fee difference between levels is a rounding error against a lot of unsellable stock.

A worked plan for a 1,200-unit first order

Concrete all the way through. You're importing 1,200 × PHONE-STAND-BLK-01 at USD 1.85 (illustrative), first order from a factory you've never shipped with. The plan:

1. Set the classes for this product, in writing. Critical: cracked acrylic leg, exposed sharp edge. Major: hinge loose enough that the stand tips under a phone's weight; non-slip base missing. Minor: small scuff on the underside. This list goes in the PO — the factory needs to know what's being counted before units exist.
2. Level III, because it's a first order: the 501–1,200 lot band → sample 125 units, selected at random from finished cartons across the whole run, not the top layer of one pallet.
3. The counters: majors at AQL 2.5, accept 7 / reject 8. Minors at AQL 4.0, accept 10 / reject 11. Criticals: any one fails the lot.
4. The report comes back: 2 majors, 8 minors, 0 criticals. Majors: 2 ≤ 7 → pass. Minors: 8 ≤ 10 → pass. Lot passes. Total inspection cost: one man-day on site, billed in the few hundred dollars per day that third-party firms typically charge (illustrative; QIMA and SGS both publish current day rates, accessed November 2026).

Change one number and watch the machinery work: if the report had shown 11 minors, the lot fails on minors — even though majors passed. The standard remedy is not you eating 1,200 units: the factory re-sorts or reworks the full lot at their cost, and a re-inspection confirms the fix before shipment. That's the leverage the sampling table buys — the failure is documented in a shared standard, not argued as opinion.

Flip the failure onto the majors — a report showing 8 majors, one past the accept number — and the same remedy applies, this time on the defect the market actually feels. Either counter tripping stops the shipment; that is the point of running them separately.

Flip it around, too: a lot can pass cleanly at 2 majors / 8 minors and still contain roughly 2–3% majors in the unsampled 1,075 units. Sampling is probability, not certification. It's the cheapest screen available per dollar, and it's not a guarantee.

The ways merchants use AQL wrong

• Treating AQL 2.5 as "2.5% of my order may be defective." The threshold applies to a sample, calibrated against process averages; a passing lot can still contain a few percent of defects. Demand refunds per the report, not per the percentage.
• Demanding zero defects at AQL. Zero-tolerance on majors sounds strong and reads to factories as inexperience; zero belongs on criticals only, where it's standard and enforceable.
• Dropping to Level I to save the inspection fee on a first order. The fee is a rounding error against the cost of receiving 1,200 unsellable units. Level III exists for exactly this order.
• Letting the factory choose the sample. Random selection from across the run is the whole protection. If the inspector samples from one staged pallet, you paid for theater.
• Treating a failed lot as a catastrophe. A failed inspection before shipment is the system working: the fix is a re-sort on the factory's floor at their cost. The expensive failure is discovering the defect rate at your own dock, with freight already paid.

When a lot does fail and the sample says so, the next decision — accept with a discount, force rework, or reject outright — has its own arithmetic, covered in the failed-shipment decision guide. And the terms that make the sampling plan enforceable get written into the PO itself, in the framing laid out in the PO-process guide.

Common questions

What does AQL 2.5 actually mean? It's the worst defect rate a sampling plan is designed to pass routinely — not "2.5% of your order can be defective." On a 1,200-unit order it means inspect 80 units and accept if the sample shows 5 or fewer majors, reject at 6. The percentage sounds looser than 2.5 because the standard screens process drift, not individual units.

How many units should I inspect on a 1,000-unit order? At the default Level II, 80 units, drawn at random from finished cartons across the run. Accept the lot on 5 or fewer major defects and reject at 6; minors run on their own counter of 7 accept / 8 reject at AQL 4.0. A first order from a new factory justifies Level III: 125 units.

Which AQL values should I use for critical, major, and minor defects? The industry convention is zero tolerance on criticals (any single one fails the lot), AQL 2.5 for majors, and AQL 4.0 for minors. Inspection companies and factories all recognize these settings, which makes reports hard to dispute. Custom values are possible but you'll spend negotiation capital better saved for the failure itself.

Does an AQL pass guarantee my order is defect-free? No. A passing lot can still contain roughly a few percent of defects in the unsampled units — sampling is probability, not certification. What the pass does mean is that the factory's process was running close to the accepted rate when the sample was drawn, and that a later dispute rests on a documented standard rather than opinion.

Who pays for re-inspection after a failed lot? When the failure is the factory's — defects against an agreed standard — the re-sort or rework happens at their cost, and the re-inspection fee is typically negotiated onto them or split. Third-party inspections bill by the man-day, usually in the few hundred dollars per day (illustrative). Put "rework at supplier cost plus re-inspection" in the PO terms before anyone disagrees.

Next step

Three numbers belong in your next PO before it ships: the defect classes for the product, the sampling level, and the AQL settings (2.5 majors / 4.0 minors / zero criticals) with the accept-reject numbers from the table above. The inspection event that executes it is laid out in the pre-shipment QC checklist — the math above and the checklist together are the whole system.