The Acceptable Quality Level (AQL) is the worst average defect rate a sampling plan treats as acceptable: the quality level with a high probability of acceptance, conventionally about 95%. It sets the accept and reject numbers for an inspection sample, so a lot at or better than the AQL almost always passes and worse lots are increasingly likely to be rejected.
AQL is how you inspect a shipment without measuring every piece. Instead of 100% inspection, you pull a statistically sized sample, count defects, and compare against an accept number the AQL defines. The idea is old and well-tested, codified in ANSI/ASQ Z1.4 and the same math explains both why it works and where it fools people. Note the vocabulary: the current standard calls it the "Acceptance Quality Limit," while "Acceptable Quality Level" is the older name for the same idea.
What does AQL actually mean?
AQL is a rate, not a target. An AQL of 1.0% does not mean you are aiming to ship 1% defective parts; it means the sampling plan is tuned so that a lot running at 1.0% defective has a high chance of being accepted, and lots meaningfully worse than that get caught more and more often. It is the dividing line the plan is built around, expressed as percent defective or defects per hundred units.
Because it is defined statistically, AQL carries an uncomfortable truth: sampling can never guarantee a lot is defect-free. A plan tuned to a 1.0% AQL will still occasionally accept a lot that is worse and occasionally reject a lot that is fine. Those two error types have names, and understanding them is the whole game.
How does AQL turn into an accept or reject number?
The chain runs from lot size to sample size to a pass/fail rule. In Z1.4 you start with the lot size and a general inspection level (Level II is the default), which gives a sample size code letter. That letter plus your chosen AQL points to a row in the master table with three numbers: the sample size, the acceptance number (Ac), and the rejection number (Re). You inspect that many pieces, count defects, and accept the lot if defects are at or below Ac, reject if they reach Re.
Z1.4 also builds in switching rules. Start on normal inspection; a run of accepted lots lets you switch to reduced inspection (smaller samples), while trouble switches you to tightened inspection (stricter accept numbers). The switching is the part people skip, and it is the part that actually protects the customer over a stream of lots rather than a single shipment.
A worked example makes the chain concrete. Say a lot of 3,200 pieces arrives and you inspect at general Level II. That lot size maps to sample size code letter L, which under normal inspection calls for a sample of 200 pieces. At a major-defect AQL of 1.0%, the table gives an acceptance number of 5 and a rejection number of 6: pull 200 pieces at random, and accept the lot if 5 or fewer are defective, reject if 6 or more turn up. The same 200-piece sample can carry a second, tighter AQL for critical defects, say 0.10% with an acceptance number of 0, and a looser one for minor cosmetic defects, so one draw of parts settles several defect classes at once. Change nothing but the lot size, and the code letter and sample size move with it; change the inspection level, and the discrimination of the plan moves with that.
Some plans go further than a single draw. Double and multiple sampling plans let a borderline first sample trigger a second (or several more) smaller samples before committing to accept or reject, which lowers the average number of pieces inspected when lots are clearly good or clearly bad. Sequential sampling takes the same idea to its limit, deciding after each piece whether the evidence is yet conclusive. All of them trace back to the same OC-curve logic; they just spend inspection effort more efficiently across a stream of lots.
What is the OC curve, and what does it show?
The operating characteristic (OC) curve is the graph that tells the truth about a sampling plan. It plots the probability of accepting a lot (vertical axis) against the lot's actual defect rate (horizontal axis). Every sampling plan has one, and it is the honest picture of what the plan will and will not catch.
Read left to right, the curve starts near 100% acceptance for very good lots and falls toward 0% for bad ones. The AQL sits high on the curve, at roughly 95% acceptance. That gap, the 5% chance of rejecting a lot that is actually at the AQL, is the producer's risk. Far down the curve sits a much worse quality level, the rejectable quality level (RQL), often set where acceptance has dropped to about 10%. The 10% chance of accepting a lot that bad is the consumer's risk.
The shape of the curve depends on the sample size, not the lot size. A bigger sample makes the curve steeper, discriminating more sharply between good and bad lots and shrinking both risks. This is why "just inspect a few pieces" gives a lazy, gently sloping curve that barely distinguishes a 1% lot from a 5% one, and why the tables scale sample size with lot size to hold discrimination roughly constant.
How do you choose the right AQL?
The AQL should track the consequence of the defect, not a round number that feels strict. Z1.4 and its users conventionally split defects into classes: critical (could cause harm or total failure), major (likely to cause a functional failure), and minor (cosmetic or unlikely to affect use). Tighter AQLs go on the classes that hurt.
- Classify the defect types for the part into critical, major, and minor, based on what each one costs the end user and your customer.
- Assign a tighter AQL to worse classes. Critical defects often carry an AQL near zero; majors a small percentage; minors a looser figure. One inspection can carry several AQLs at once.
- Pick the inspection level. Level II is the default; drop to Level I when inspection is expensive and the risk is low, raise to Level III when discrimination matters more than cost.
- Read the plan from the tables. Lot size and level give the code letter; the letter and each AQL give a sample size and accept/reject numbers.
- Honor the switching rules. Move to tightened inspection when quality slips and to reduced inspection when a run of good lots earns it. Skipping this quietly breaks the protection.
How does AQL relate to SPC and process control?
AQL sorts lots after the fact; statistical process control keeps the process from making bad lots in the first place. They answer different questions. Acceptance sampling asks "should I take this shipment?" while SPC and control charts ask "is my process stable and centered right now?" A mature quality system leans on SPC and shrinks incoming inspection over time, because a process proven capable through capability studies makes lot-by-lot gatekeeping largely redundant. AQL sampling is a safety net; capable processes are the floor.
AQL also depends on the kind of data you collect. Z1.4 governs attribute inspection where each piece is judged pass or fail; its sister standard, Z1.9, handles variable data where you measure a dimension. Attribute sampling needs larger samples for the same discrimination, which is the practical cost of a simple go/no-go check.
What are AQL's limits?
The biggest trap is reading AQL as a quality goal. Suppliers who "hit the 2.5% AQL" every month are not being asked to make 2.5% junk; they are passing a gate that tolerates it, and a supplier optimizing to the gate rather than to zero defects is a supplier quality problem waiting to surface. Sampling also cannot find a defect that is not in the sample, so it is weak against rare, high-severity defects, exactly the ones that warrant tighter AQLs or 100% inspection. And every accepted lot that later fails becomes a nonconformance and a hit to your cost of quality which is the real scoreboard behind the sampling math. The standards facts worth keeping straight:
- ANSI/ASQ Z1.4 provides attribute sampling plans indexed by AQL, with normal, tightened, and reduced inspection and switching rules for a continuing stream of lots (ASQ, Z1.4 & Z1.9).
- The AQL corresponds to a high probability of acceptance on the plan's OC curve, conventionally about 95%, defining the producer's risk (ASQ).
- Discrimination between good and bad lots is driven by sample size, not lot size, which is why the tables scale samples with lot size (ASQ).
Used with eyes open, AQL is a rational way to spend inspection where it pays. Used as a target, it becomes an excuse. The practical work, keeping defect classifications, sampling plans, and disposition records straight across every supplier and lot, is the kind of paperwork Harmony's paperwork digitization and AI search pulls into one searchable place, next to your QMS rather than on top of it. See it on a plant like yours in the CLS case study.