AQL sampling explained: how many to check and when to reject

What an AQL sampling plan does

Checking every unit in a shipment is slow, expensive and, for some tests, impossible. Sampling solves that. You inspect a set number of units drawn at random from the lot, count the defects you find, and compare the count with a pre-agreed accept and reject number.

An AQL plan gives you three numbers:

• Sample size (n): how many units the inspector pulls.
• Accept number (Ac): the most defects the sample can contain and still pass.
• Reject number (Re): the count at which the lot fails. In single sampling, Re is always Ac + 1.

The plan comes from two tables in ANSI/ASQ Z1.4 (the US successor to MIL-STD-105E) and its international twin, ISO 2859-1. Table I turns lot size and inspection level into a code letter. Table II turns the code letter and your AQL into the sample size and the Ac and Re numbers. Our free AQL sampling planner reads both tables for you, including the arrow rules.

Step 1: pick the inspection level

The inspection level controls how big the sample is relative to the lot. A higher level means a larger sample and better discrimination between good and bad lots.

In practice, most pre-shipment inspections in India and China run at General II for visual and workmanship checks, with a special level for on-site tests. A carton drop test at S-1 or S-2, for example, sacrifices only a few cartons. For carton testing in a lab, see the ISTA 1A drop test planner.

Step 2: find the code letter

Table I of Z1.4 maps the lot size to a code letter. At General Level II the bands are:

Notice how slowly the sample grows. A lot of 3,201 units and a lot of 10,000 units both get a sample of 200. That is why splitting one shipment into several small “lots” for inspection costs more inspector days without improving the decision much.

Step 3: set an AQL for each defect class

Defects are classed by their effect, and each class gets its own AQL:

• Critical: could harm the user or breach a legal requirement. A sharp edge on a child’s product, a missing earth connection, a wrong safety warning.
• Major: likely to cause a return or make the product unfit for use. A lid that does not seal, a seam that opens, a dimension out of tolerance on a fit feature.
• Minor: a departure from the specification that most customers would accept. A small scuff on a non-visible face, a loose thread end.

A common consumer-goods setting is zero acceptance for critical defects, AQL 2.5 for major and AQL 4.0 for minor. Tighter products, such as medical accessories or precision parts, use lower AQLs for majors, such as 1.0 or 0.65. The right choice depends on the cost of a defect reaching your customer, not on habit.

Worked example: a 3,500-unit lot

Two lessons from this example come up on almost every programme. First, classes are judged separately: a comfortable major count does not rescue the minor count. Second, a borderline minor count usually points to a process issue, here thread trimming and handling, that a single in-line check at 20 to 30% of production would have caught early.

What AQL does not mean

The most common misunderstanding is reading AQL 2.5 as “up to 2.5% defects are fine”. It is not a promise about the lot.

The AQL is the worst process average the plan is built to accept most of the time. A lot running exactly at the AQL will usually pass. A lot running well above it will usually fail. In between is a grey zone where the result depends on which units land in the sample. The operating characteristic curve for each plan shows those probabilities. The AQL sampling planner calculates them with a binomial model, so you can see how likely a lot at 3%, 5% or 8% defective is to pass.

This is why AQL works best as one layer of control rather than the only one. Good suppliers hold capability on key dimensions (a Cpk of 1.33 or better is a common target), run their own final checks, and use your inspection as confirmation.

Switching between normal, tightened and reduced

Z1.4 is designed to run over a series of lots, not a single one. It has switching rules:

• Normal to tightened: when two of five consecutive lots are rejected on original inspection.
• Tightened to normal: when five consecutive lots are accepted under tightened inspection.
• Normal to reduced: allowed only when the standard’s conditions are met, including a run of accepted lots, steady production and approval by the responsible authority.

Tightened inspection usually keeps the sample size and lowers the accept numbers. Reduced inspection uses smaller samples. ISO 2859-1 and Z1.4 share the normal and tightened single plans, but some reduced-inspection numbers differ, so name the standard in your contract. Many importers never use switching and run every lot at normal inspection. That is simpler, but you lose the pressure tightened inspection puts on a supplier whose quality is slipping.

Five mistakes that weaken a sampling plan

1. Sampling from one carton. Draw from across the lot. A common practice is to open a number of cartons roughly equal to the square root of the carton count, then pull units evenly from each.
2. No agreed defect definitions. “Scratch” means nothing without a size, a face and a viewing distance.
3. Inspecting too early. Pre-shipment inspection needs finished, packed goods. See the pre-shipment inspection checklist.
4. Counting defective units instead of defects, without saying so. One shirt can carry three defects. Agree which you count.
5. Treating a pass as proof of zero defects. A pass means the sample met the rule. Keep your returns data and adjust AQLs when it tells you something.