Inspection starts with your own line. Check your first 50 to 100 parts, note the spread, then put that defect rate and the cost of a miss on the table before choosing a level. A supplier that accepts a stated sample size, plan and defect class, backed by a CMM dimensional report issued against the mold number, beats the lowest AQL on paper.
Key takeaways
- ISO 2859-1 is the sampling standard behind almost every AQL table; an AQL is expressed as defects per hundred units, not a percentage of rejects.
- Use general levels I, II and III for routine work and the special S levels only where the test is destructive, slow or expensive.
- Sample size follows production quantity and inspection level, not the criticality of the part. Criticality decides the AQL, not the n.
- Critical defects consume zero in most plans, but “zero” still needs a defined sample size before it means anything.
- Reject on criteria, and give the lot a documented disposition: accept, reject, sort, rework or concession.
- ±0.01 mm on critical molded dimensions is a toolbox limit, not a tolerance you can inspect at AQL without a measurement plan attached.
What AQL actually measures
AQL stands for acceptance quality limit. It is the worst quality level, measured in percent nonconforming, that is still acceptable as a long-run average for a lot submitted for inspection. That definition is doing a lot of work, because it is a long-run average, not a guarantee about the shipment in front of you.
Two consequences follow. First, a lot at the AQL limit will sometimes fail and sometimes pass, by design; the standard sets the risk, not a promise. Second, AQL does not describe a part. It describes an inspection plan. A drawing calling out “AQL 1.0” without a sample size, an inspection level and a defect class has not specified anything you can execute.
A common share of a molding contract is settled by quoting an AQL number and leaving the rest to habit. That is where disputes come from later, because two people can both say “AQL 1.0” and mean plans with sample sizes that differ by a factor of five.
The three words most AQL arguments leave out
Lot size. The sample size is looked up from the production quantity, so a plan written for a 2,000-piece shipment is meaningless on a 300-piece pilot run.
Inspection level. General levels I, II and III set how much of the lot you look at. Level II is the default in most purchase orders. Level I halves the burden and suits stable, well-understood processes; level III suits a new tool, a new resin or a high-cost failure.
Defect class. Critical, major and minor are defined by the customer, not by the standard, and each class gets its own AQL. A part with the wrong colour and a part with an unsafe latch should never share a sampling plan.
The standard tables: single sampling for normal inspection
The rows below are the general inspection levels at normal severity, single sampling, pulled from ISO 2859-1. Each cell is the sample size from the lot, followed by the acceptance number and the rejection number: Ac 0 / Re 1 means the lot passes only with zero nonconforming units in the sample.
| Lot size (pcs) | Level I | Level II | Level III |
|---|---|---|---|
| 151–280 | 5 (Ac 0 / Re 1) | 32 (Ac 1 / Re 2) | 50 (Ac 2 / Re 3) |
| 281–500 | 8 (Ac 0 / Re 1) | 50 (Ac 1 / Re 2) | 80 (Ac 3 / Re 4) |
| 501–1,200 | 13 (Ac 0 / Re 1) | 80 (Ac 2 / Re 3) | 125 (Ac 5 / Re 6) |
| 1,201–3,200 | 20 (Ac 0 / Re 1) | 125 (Ac 3 / Re 4) | 200 (Ac 7 / Re 8) |
| 3,201–10,000 | 32 (Ac 1 / Re 2) | 200 (Ac 5 / Re 6) | 315 (Ac 10 / Re 11) |
| 10,001–35,000 | 50 (Ac 1 / Re 2) | 315 (Ac 7 / Re 8) | 500 (Ac 14 / Re 15) |
| 35,001–150,000 | 80 (Ac 2 / Re 3) | 500 (Ac 10 / Re 11) | 800 (Ac 21 / Re 22) |
Two readings matter. Level III checks about four times the sample of level I for roughly the same lot, which is why the default level should be a deliberate choice rather than whatever the form already said. And the acceptance numbers climb quickly at the top of the table: a 20,000-piece lot at level II allows ten nonconforming units in a 315-piece sample and still passes.
Choosing the AQL for each defect class
Work the classes from the failure consequence, not from tradition. A cosmetic blemish at the back of a housing and a stressed latch on the same part deserve different plans, and the drawing should say so.
| Defect class | Example on a molded part | Working AQL range | Typical plan |
|---|---|---|---|
| Critical | Wrong resin, shuts off a safety function, sharp edge | 0.065 or zero | 100% check or a defined sample with Ac 0 |
| Major, functional | Dimension out of tolerance, a crack, a blocked boss | 0.65–1.0 | Level II on the major class only |
| Major, cosmetic | Sink on a visible face, splay, a flow mark | 1.0–1.5 | Level II, graded against an approved limit sample |
| Minor | Faint handling mark, gate vestige inside the drawing limit | 2.5–4.0 | Level I or II |
The cosmetic row is the one that generates most reject arguments. Sink, splay and flow marks are judgement calls unless a limit sample or a texture reference is signed at T1 and kept by both sides. Without that sample, “major cosmetic” is an opinion, and an AQL applied to an opinion measures nothing.
Variables data: where AQL is the wrong tool
Attribute sampling counts defects. Dimensional inspection measures them. They answer different questions, and on a tolerance as tight as ±0.01 mm on a critical molded dimension the counted answer is not good enough.
For measured characteristics the correct instrument is a capability study on the actual gauge. Take at least 30 parts across several shots, run the Cpk calculation, and compare it against gauge repeatability and reproducibility. A characteristic with a Cpk of 1.33 and a gauge R&R of 30% is not controlled; the gauge is doing the deciding.
This is where a supplier’s document set separates itself. A CMM dimensional report on trial samples says what the tool produced at T1. A capability number says what it will keep producing on a Tuesday in month six. Ask for both, and ask which dimensions sit on the capability list.
Writing an inspection clause that survives a dispute
Five lines fix most of what goes wrong. Name the standard and the edition. Name the inspection level and severity. Name the sample size or the rule that produces it, and the randomisation method. Name the acceptance criteria per defect class, with a limit sample attached for anything cosmetic. Name the disposition for a rejected lot, including who pays the sort.
Add a sentence on when the clock starts. Inspection at your dock has to happen before the lot is consumed, or a rejected lot has already been assembled into finished goods and the argument moves to who eats the scrap. Agreeing a hold period at order confirmation costs nothing and settles the question in advance.
How inspection scales with volume
Sampling plans are sized to be affordable at volume, and they change shape as the run gets longer. The pattern below is the one we see on molded parts, where tool condition rather than operator variation drives most rejections.
| Volume | What to check | Cadence |
|---|---|---|
| T1 trial, 50–200 pcs | Every drawing dimension on a CMM, full cosmetic against the limit sample | One report per trial, per mold number |
| Pilot, 1,000–5,000 pcs | Critical dimensions on a sample, cosmetic on all | First-off plus every 2 hours |
| Production | Critical dimensions per shift, cosmetic per lot, attribute plan at level II | Per AQL plan, with a first-off after any tool stop |
| After a tool repair | Critical dimensions and the repaired feature | First-off before the run restarts |
The third row is where attribute sampling earns its keep, and the fourth is where most escapes happen. A tool that has been apart and back together is a different tool until the first-off proves otherwise.
What we hand over, and what we do not sign
Our answer: we will write the plan with you at the quotation stage, because the inspection level changes the price per part and hiding that until after T1 wastes everyone’s month. On a new tool we inspect the part, not the shipment, and we say which dimensions are capability-critical.
RCH Plastic reference: we machine molds in-house to ±0.005 mm on critical mold dimensions, through 13 CNC machining centres with travels up to 1,600 mm, six EDM machines including Sodick mirror-finish and twin-head units, and three wire EDMs. That tool can then hold ±0.01 mm on critical molded dimensions, subject to the resin and the part geometry. Product certification is not how we evidence quality; the evidence is documentary, issued against your mold number.
The document set is samples, trial photos, a trial video, the full process parameter sheet and a CMM dimensional report on trial samples. Every claim in this article is backed by one of those, traceable by mold number, which is also how we close an inspection dispute without re-arguing the standard.
Related guides in this series
- Defects and quality guide — the nine defects worth rejecting a lot over, and what causes each
- Injection molding tolerances — how to read a tolerance before you decide what to inspect
- Common molding defects — root causes you can trace back to the tool or the process
- Mold surface finish — turning a cosmetic specification into a grade both sides can check
- Mold flow analysis — the fill, pack and cooling results behind a trial report
- Mold manufacturing FAQ — lead time, life and how dimensional precision is verified
- Connector housing case study — 8 cavities held inside a ±0.01 mm class across the tool
- How to get a mold quote — where the inspection clause belongs in the RFQ
Put the plan in the RFQ, not in the argument afterwards
Send your drawing with the defect classes marked and we will come back with the sample size, the level and the dimensions worth putting on a capability list, together with a DFM report within three working days at no cost. Request a free DFM review and bring the inspection clause to the same meeting.
