AQL sampling explained: levels, tables and a worked lot.
You receive a lot of 3,200 machined parts. You cannot measure all of them, but you must decide accept or reject. AQL sampling, codified in ISO 2859-1 and ANSI/ASQ Z1.4, tells you exactly how many to inspect and how many defects you are allowed to find. Here is how the system works, end to end.
What AQL actually means
AQL stands for Acceptable Quality Level. It is the maximum percent defective that is still treated as an acceptable process average. An AQL of 1.0 means a supplier whose long-run process runs at 1.0 percent defective should see almost all of their lots accepted. It is a property of the sampling plan, not a target you give the operator and not a promise that every accepted lot is below that level.
Three numbers drive every plan: the lot size, the inspection level, and the AQL you agree with the customer. From those three you derive a sample size and an acceptance number. That is the whole machine.
How a sampling plan is built
The standard turns your three inputs into a plan in two lookups:
- Lot size + inspection level → code letter. A table maps the lot-size range and the chosen level to a single letter, A through R.
- Code letter + AQL → sample size, Ac, Re. The single sampling table gives the sample size n, the acceptance number Ac, and the reject number Re for that letter and AQL.
You then pull a random sample of n pieces, inspect them, and count defectives. If defectives are at or below Ac, accept the lot. If they reach Re (always Ac plus one), reject it.
General and special inspection levels
The inspection level decides how hard you look relative to lot size. ISO 2859-1 gives three general levels and four special levels:
| Level | Use it when | Sample vs lot |
|---|---|---|
| General I | Defects are cheap, you want lower inspection cost | Smaller |
| General II (default) | Most situations, the standard starting point | Medium |
| General III | Defects are costly, you want tighter discrimination | Larger |
| Special S-1 to S-4 | Destructive or expensive tests where small samples are unavoidable | Very small |
If nobody specifies a level, use General Level II. Move to Level III for safety-critical or warranty-sensitive parts, and to Level I only when the cost of an escaped defect is genuinely low.
Reading the code letter
The first table converts lot size and level into a letter. A few anchor points for General Level II:
| Lot size | Code letter (Level II) |
|---|---|
| 91 to 150 | F |
| 151 to 280 | G |
| 281 to 500 | H |
| 501 to 1,200 | J |
| 1,201 to 3,200 | K |
| 3,201 to 10,000 | L |
The letter is just an index into the second table. It does not by itself tell you the sample size until you combine it with the AQL.
A worked example: 3,200-piece lot
Suppose a Pune CNC shop ships a lot of 3,200 turned shafts. The customer agreement is General Level II, single sampling, normal inspection, with AQL 1.0 for major defects and AQL 2.5 for minor defects.
- Code letter. Lot 3,200 at Level II lands at the top of the 1,201 to 3,200 band, giving code letter K.
- Sample size. Code letter K corresponds to a sample of 125 pieces.
- Acceptance numbers. At AQL 1.0, the K row gives Ac = 3, Re = 4. At AQL 2.5, it gives Ac = 7, Re = 8.
So you pull 125 random shafts. If you find 3 or fewer major defects, the lot passes on majors; 4 rejects it. Separately, 7 or fewer minor defects passes on minors; 8 rejects. A lot can pass minors but fail majors, or the reverse — each class is judged on its own line.
Normal, tightened and reduced inspection
AQL sampling is not static. The switching rules move you between three regimes based on recent history, which is what gives the scheme its teeth:
- Normal: the default starting point.
- Tightened: triggered when 2 of 5 consecutive lots are rejected. Sample sizes stay similar but acceptance numbers drop, so it is harder to pass. If 5 consecutive tightened lots pass, you switch back to normal.
- Reduced: allowed after a run of good lots under defined conditions. Smaller samples lower your cost, rewarding a proven supplier.
Common mistakes
- Treating AQL as a pass mark per lot. AQL describes a process average and producer risk, not a guarantee about any single accepted lot.
- Non-random sampling. Pulling the top tray or the easy-to-reach parts breaks the statistics. The 125 must be drawn at random across the lot.
- Wrong defect classification. Mixing a critical defect into the major line understates risk. Critical defects usually run at a far tighter AQL, often 0 acceptance.
- Ignoring switching rules. As above, staying on normal forever defeats the purpose.
- Forgetting that sampling cannot improve a lot. If the process is bad, sampling just decides accept or reject. Fixing the process is a separate job — capability studies and FMEA, not more inspection.
Where ballooning and inspection meet
Sampling tells you how many parts to inspect; the inspection itself still needs a numbered, traceable characteristic list for each part you pull. If you are inspecting against a drawing, balloon every characteristic first so each measured feature ties back to a number on the report. CadNexa's auto-ballooning tool detects dimensions and GD&T frames on a PDF drawing and exports a numbered inspection sheet, which pairs neatly with an AQL plan for incoming or in-process inspection.
For related reading, see our guides to Cp and Cpk for judging whether the process behind the lot is even capable, and PPAP for Indian OEMs for where sampling plans sit inside a full submission. Ready-to-use forms are on the templates page.