An inspector walks into a factory in Dongguan, opens the cartons holding a finished run of 5,000 units, and pulls 200 of them at random. If 10 or fewer have a major defect, your order ships. If 11 do, the whole lot is on hold. That single pass-or-fail line is set by a standard called AQL, and most importers find out how it works only after a shipment they already paid for gets rejected, or worse, after a bad one gets waved through.
AQL stands for Acceptable Quality Limit. The name is misleading, because it does not describe a quality you want. It describes the most defects a batch can contain before you refuse it. Set it loosely and the factory ships you goods you would have rejected by hand. Set it without telling the factory in writing, and the inspector and the supplier are working from two different rulebooks on the day the goods are sitting on the floor.
This is the standard behind every pre-shipment inspection booked from China, and the three numbers inside it are yours to set before production. Most importers let the factory set them instead, which is the first place money leaks.
What AQL actually measures
AQL is defined in two near-identical standards: ISO 2859-1, published by the International Organization for Standardization, and ANSI/ASQ Z1.4, its American twin from the American Society for Quality. Both descend from a US military sampling standard and both use the same tables, so an inspection agency in Ningbo and a buyer in Toronto can read the result the same way.
The standard does one job. It tells an inspector how many units to pull from a finished lot and how many defects in that sample are allowed before the lot fails. You inspect a sample instead of all 5,000 pieces because checking every unit is slow, expensive, and on many products destructive. A statistically chosen sample gives you a defensible read on the whole batch.
Before any of that works, defects get sorted into three classes, and this is where most of the real decisions live. A critical defect makes the product unsafe or illegal to sell: a sharp edge on a child's toy, a mains cable with no earth, a cosmetic that leaks. A major defect is a fault a normal customer would notice and return the product for, like a dead button, a broken zip, or the wrong colour. A minor defect is a small flaw that does not affect function and that most buyers would accept, such as a loose thread or a faint scuff on the base. You assign a separate AQL to each class, and the three numbers do not have to match.
The three numbers you set before production
For consumer goods, the standard combination is 0 for critical defects, 2.5 for major, and 4.0 for minor. The number is the percentage of the lot you will tolerate in that class. A major AQL of 2.5 means you accept a batch whose major-defect rate sits at or below 2.5 percent. AQL 2.5 is the most widely used setting for major defects in consumer goods worldwide, and 4.0 is the common ceiling for minor ones.
Critical defects are the line you do not move. Setting critical to 0 means a single unsafe unit in the sample fails the lot. For a product that goes near a child, an electrical socket, or skin, that is the only setting that makes sense, and it is the one that keeps a product recall off your books later.
The numbers are a business decision, not a technical one. A promotional giveaway can run looser. A premium product sold under your own brand, where one bad review costs you fifty sales, runs tighter: major 1.0, minor 2.5. The factory will often propose its own AQL, and its proposal will be more generous than yours, because a looser limit means more of its output passes. Whatever you choose, it belongs in the purchase order and the contract, in writing, agreed before the first unit is made. An AQL agreed after the goods are boxed is an AQL the supplier can argue with.
How the sample size is chosen
You do not pick the sample size. The standard does, from two inputs: the size of the production lot and an inspection level.
Most importers use General Inspection Level II, the default for consumer products. Level I pulls a smaller sample and costs less but catches less. Level III pulls a larger sample for goods where you want more confidence. Level II sits in the middle and is what an inspection agency will assume unless you say otherwise.
The lot size and the level point to a code letter, and the code letter points to a sample size. A run between 3,201 and 10,000 units maps to code letter L, which sets a sample of 200. Take that 200-unit sample at a major AQL of 2.5 and the table gives you an acceptance number of 10 and a rejection number of 11. Ten or fewer major defects in the 200 you checked, the lot ships. Eleven or more, it fails. At the minor AQL of 4.0, the same 200-unit sample accepts up to 14 and rejects at 15.
A smaller run scales the same way. A 2,000-unit order maps to code letter K and a 125-unit sample, where a major AQL of 2.5 accepts up to 7 defects and rejects at 8. Notice the sample barely grows as the lot does. A 2,000-unit run and a 10,000-unit run are checked with 125 and 200 units, not numbers ten times apart, because the statistics that make sampling work do not scale linearly with lot size. Importers who expect a 50,000-piece order to be inspected one unit at a time are reading the wrong standard.
That is the entire mechanism. Lot size sets the sample, the AQL sets the accept-or-reject count, and the inspector pulls units at random across the cartons rather than from the one pallet the factory pushes forward. The randomness matters as much as the math. A sample drawn only from the cartons nearest the door tells you about those cartons and nothing else.
The point of the standard is that everyone reads the same result. Your inspector in Shenzhen, your supplier, and you in another country are all looking at one number against one threshold you all agreed to in advance.
Where importers get burned
The math is the easy part. The losses come from four habits.
The first is letting the factory set the AQL. If you never specify a limit, the inspection runs against the supplier's, and the supplier's is loose by design. You inherit a defect ceiling you never agreed to and find out only when the goods arrive. Setting the limit yourself is a separate job from vetting the supplier: one checks the factory, the other checks the goods that factory made.
The second is skipping the defect classification. AQL numbers mean nothing until critical, major, and minor are defined for your specific product, with photos of what each looks like. Without a written defect list, the inspector and the factory argue on the day about whether a mark is major or minor, and the argument happens while your balance payment is already due.
The third is timing. A pre-shipment inspection is run when roughly 80 to 100 percent of the order is produced and packed, before you pay the balance and before the goods leave the factory. That sequence is the whole point. It is the last moment the advantage sits with you, because the supplier still wants the final payment. Book the inspection after you have wired the balance and a failed lot becomes the factory's problem to fix on its own schedule, which means yours. This is also why the inspection has to sit inside your payment terms, not after them. The same logic applies to your shipping terms: under DDP or any arrangement where you pay on arrival, a lot you cannot reject at the factory is one you are stuck with at port.
The fourth is the ship-anyway call. A lot fails, the vessel cuts off in three days, and the pressure to load it regardless is real. Sometimes a waiver on a borderline minor result is the right commercial choice. Making that call blind, without an inspection report in front of you, is how a critical defect reaches a customer and turns into a return, a chargeback, or a regulator's letter.
A failed lot is not a dead order. The factory sorts the batch, reworks or replaces the defective units, and the reworked lot gets inspected again against the same AQL before it ships. What decides who pays for that second inspection and the rework is whatever your contract says, which is one more reason the AQL and the inspection belong in the agreement rather than in a verbal understanding.
The fix for all four is the same and it is unglamorous. Agree the AQL and the defect classes in writing before production. Book the inspection before the balance payment. Pull the sample at random. Read the report before you release the goods. Mila's sourcing agents run a three-stage QC on the ground in China and book the pre-shipment inspection into the payment schedule. The full report, photos and accept-or-reject result land in the same WhatsApp thread you use to run the order, so the ship-or-hold decision stays yours and you make it with the numbers in front of you. If a China order is in production now, audit your current supplier in 48 hours and have an agent confirm the AQL against what is actually in the cartons.
Sources: ISO 2859-1:1999, Sampling procedures for inspection by attributes, Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot inspection; ANSI/ASQ Z1.4, Sampling Procedures and Tables for Inspection by Attributes (American Society for Quality).