What you need before you start
- Lot size N: the number of units in the lot you are judging, not the order quantity.
- Inspection level: general level II is the usual default.
- AQL: agreed with the customer or set in your control plan, often different for critical, major and minor defects.
- Inspection state: normal, tightened or reduced. This page uses single sampling, normal inspection.
The tables themselves are in ISO 2859-1 and ANSI/ASQ Z1.4 (both based on MIL-STD-105E). We do not reproduce them here; the AQL calculator looks up the same values.
Worked example: lot of 1,200 housings
- Step 1, code letter: lot size 1,200 falls in the 501 to 1,200 band. At level II that is letter J.
- Step 2, sample size: letter J means n = 80.
- Step 3, Ac/Re: in the single-sampling normal table, row J, column AQL 1.0 gives Ac 2, Re 3. No arrow, so the plan stands.
- Step 4, decision: take 80 housings at random from across the lot. Find 0, 1 or 2 nonconforming: accept. Find 3 or more: reject.
Same lot, tighter AQL 0.65: still letter J and n = 80, but Ac 1, Re 2. Looser AQL 2.5: Ac 5, Re 6. One lot more, 1,201 units, moves you to letter K and n = 125 (at AQL 1.0: Ac 3, Re 4). A small lot of 150 at AQL 0.65 is letter F, n = 20, Ac 0, Re 1.
What that plan actually accepts
A sampling plan is a filter with a probability curve (the OC curve), not a gate. For n = 80, the chance of accepting a lot depends on its true defect rate. Computed with the binomial model, per 1,000 lots:
| True defect rate in the lot (ppm) | Lots accepted per 1,000, Ac 2 (AQL 1.0) | Lots accepted per 1,000, Ac 1 (AQL 0.65) |
|---|---|---|
| 5,000 (1 in 200) | 992 | 939 |
| 10,000 (1 in 100) | 953 | 809 |
| 20,000 (1 in 50) | 784 | 523 |
| 30,000 (3 in 100) | 568 | 304 |
| 50,000 (1 in 20) | 231 | 86 |
Read that twice. A lot with 1 defective in 50 (20,000 ppm, double the AQL) still passes the AQL 1.0 plan about 784 times in 1,000. A lot with 1 in 100 passes about 953 times in 1,000. If your customer measures you in single-digit ppm, AQL sampling at incoming will not protect them, and it will not find your problem quickly either.
Common mistakes
- Using the order quantity, or the whole shipment of several lots, as N.
- Taking the sample from the top of one box instead of across the lot.
- Ignoring the arrow and keeping the original letter's sample size.
- Counting defects per part when the plan is for nonconforming items (or the reverse).
- Reading “accepted” as “zero defects”. An Ac 0 plan with n = 20 still accepts a 1-in-20 lot about 358 times in 1,000 (about, binomial approximation).
- Never applying the switching rules to tightened inspection after rejections.
AQL in an 8D: where it fits and where it does not
During D3 containment, a known defect is in the population. Sampling only tells you it is probably still there. Sort 100 percent with a clear sorting instruction, record the counts, and use a sampling check afterwards only to verify the sort, if the customer agrees. In D6, a run of accepted lots is weak evidence on its own; show the defect rate before and after over a comparable volume (see D6 verification). For process-level evidence, capability says more than sampling: see Cpk vs Ppk.
FAQ
- Does AQL 1.0 mean 1 in 100 defective parts is acceptable?
- No. AQL is a property of the sampling scheme over a series of lots: the quality level that the plan accepts most of the time. It is not a permission to ship defects. Many automotive customers expect zero defects or set ppm targets regardless of the sampling plan; read your customer's requirements.
- What does Ac and Re mean?
- Ac is the acceptance number: accept the lot if the number of nonconforming items in the sample is Ac or fewer. Re is the rejection number: reject the lot if it is Re or more. In single sampling Re is always Ac + 1.
- What is the arrow rule?
- Some cells in the plan table hold an arrow instead of Ac/Re. Follow the arrow to the first cell with numbers and use that row's sample size, not your original code letter's. The AQL calculator applies this automatically.
- Which inspection level should I use?
- General level II unless your customer agreement or control plan says otherwise. Level I takes smaller samples and discriminates less; level III takes larger samples. Special levels S-1 to S-4 are for destructive or expensive tests.
- Can I use AQL sampling to release a suspect lot during containment?
- Usually not on its own. When a known defect is in the population, the customer normally expects 100 percent sorting (or a validated automatic check) until the cause is fixed. AQL sampling suits routine receiving inspection, or a check on already-sorted stock if the customer agrees.