AQL Sampling for Garment Inspection — ISO 2859-1 Made Practical

Last month our buyer's QC team arrived for a shipment inspection. 4,000 T-shirts packed in 200 cartons. They opened five cartons at random, pulled out 200 garments, and started inspecting. By the end of the afternoon they had found 9 major defects, accepted the lot, and the container left.

Why 200 garments? Why was 9 defects okay but 11 would have been a reject? Why did they not just inspect 100? Or 500?

The answer is AQL sampling, defined by ISO 2859-1:1999. It is the universal language of garment inspection — used by QIMA, SGS, Bureau Veritas, Asia Inspection, every brand's internal QC team, and every CMT factory's outgoing inspection desk. If you ship to brands, your shipment is inspected this way. If you do not understand how, you cannot defend or argue an inspection result.

This guide is the version I wish someone had handed me when we started exporting. Worked examples, no jargon, with a free online calculator at the bottom.

What AQL actually means

AQL stands for Acceptable Quality Limit. It is not a target — it is a statistical threshold. AQL 2.5 does not mean "I want 2.5% defects". It means "I am willing to accept lots that have up to a 2.5% defect rate in their long-run average."

The key insight: you cannot inspect 100% of the shipment — that is too expensive and time-consuming. Instead, you pull a small random sample, count defects, and use a statistical table to decide accept or reject. The table is calibrated so that lots near the AQL get accepted most of the time, and lots far above the AQL get rejected most of the time.

The two tables you need

ISO 2859-1 gives you two lookup tables. Together they answer the only two questions that matter: how many to sample, and how many defects trigger a reject.

Table I — Sample size code letters

Table I (a.k.a. Table A) maps your lot size and your inspection level to a letter (A through R). The letter is just an index into Table II.

Default inspection level for garment buyer QC: General II. Use General III only when the buyer is being aggressive (premium brands, repeat quality failures). Use General I for low-risk products or repeat-good-supplier reduced inspection.

Table II — Sample size + accept/reject numbers

Table II takes the letter from Table I and your chosen AQL, and gives you two numbers: how many to inspect, and the accept/reject thresholds.

Example: Lot 2,400, General II → Letter K. At AQL 2.5, K means sample 125 garments, accept if defects ≤ 7, reject if defects ≥ 8.

If you cross-check that example against QIMA's published reference table, it matches. (We tested all values in our open-source garment-aql-calculator against QIMA's reference — that is how we caught and fixed a transcription bug in our initial implementation.)

The three AQL tiers — critical, major, minor

This is where most factory QC supervisors get confused. Every garment inspection runs three AQL plans simultaneously, not one.

So for the 2,400-piece lot at General II, the inspector runs all three plans on the same 125-garment sample:

The lot is accepted only if all three tiers pass. One critical defect over the threshold = full lot reject, even if the major and minor tiers are clean.

Why the lot rejected even though only 8 defects were found

A common confusion in our factory: "But supervisor, we only found 8 defects out of 125 garments — that is 6.4%, less than the 2.5 AQL we agreed to. Why did the buyer reject?"

Because AQL is not a percentage threshold for the sample. It is the long-run average defect rate that the sampling plan is calibrated to accept. The accept number (7 in this case) is set by statistical math so that lots actually at 2.5% defect rate get accepted 95% of the time — not because 7/125 = 5.6% somehow equals "2.5 AQL".

If you want to argue an inspection result, the only thing to argue is whether the inspector correctly classified the defects (Major vs Minor — there is judgment there). Arguing the math is pointless.

A worked example — 4,000-piece T-shirt shipment

Real scenario from our factory last month. 4,000 T-shirts, ready to ship to a US brand. Buyer specified General II / AQL 0.65 critical / 2.5 major / 4.0 minor.

Step 1. Find the code letter. Lot 4,000 + General II → Letter L.

Step 2. Look up L in Table II for each AQL:

Step 3. Buyer pulls 200 random garments. Finds: 0 critical defects, 9 major defects, 6 minor defects.

Step 4. Check each tier:

• Critical: 0 ≤ 3 ✓ pass
• Major: 9 ≤ 10 ✓ pass
• Minor: 6 ≤ 14 ✓ pass

Lot accepted. Container ships.

The cost of inspection — and why you should pre-inspect

Buyer inspection happens after you have built and packed the entire lot. If the inspector rejects, you re-inspect 100%, sort the defective pieces out, repack, and the buyer either accepts the sorted-and-repaired lot or demands a discount (often 5–15% off invoice value). Either way, your margin is destroyed.

The fix is internal pre-inspection using the same AQL plan the buyer will use. We run AQL inspection at three points:

1. During production — every 250 pieces from the line gets a 20-piece spot inspection. If we find more than 1 major defect, line manager investigates the operator/operation. This catches problems before they propagate.
2. At finishing — every full bundle gets an AQL inspection before going to packing. This is the gate that 100% of garments pass through.
3. Pre-shipment — 24 hours before the buyer inspector arrives, we run an internal pre-shipment inspection at the same AQL the buyer will use. If we fail our own pre-inspection, we have time to sort.

Three internal AQL inspections, then one buyer inspection. By the time the buyer arrives, we usually know whether it will pass — and if we are unsure, we delay the inspection and fix the issues first.

Common mistakes

1. Treating AQL like a target

AQL 2.5 is the limit you are willing to accept. Your actual production should aim for much lower — ideally a major DHU under 5% (see our DHU reduction case study).

2. Mixing up sample size with lot size

"We have 5,000 pieces, so we need to inspect 5%". No. ISO 2859 sample sizes are absolute, not percentages. A 5,000-piece lot at General II uses 200 garments — that is 4%. A 50,000-piece lot also uses 315 garments at General II — that is 0.6%. The math is statistical, not proportional.

3. Inspecting non-random samples

The inspector pulls cartons in a specific random pattern — first carton, then every Nth — to avoid bias. If your packer puts the worst pieces at the bottom of the carton, the inspector will find them. Random sampling only works if your defect distribution is random.

4. Arguing about classification

This is the only place worth arguing. The buyer's published defect classification list (what counts as Major vs Minor) is part of the PO. If an inspector classifies "uneven topstitch" as Major when your buyer's spec list calls it Minor, push back with the spec sheet. We have saved at least three rejections this way.

What about Special Inspection Levels (S1, S2, S3, S4)?

You will see these on some buyer PO sheets. They are for inspections that are destructive (you cut the garment open to check seam strength), expensive (lab tests for shrinkage, colour fastness), or slow (full measurement of every dimension). They use much smaller samples than General levels.

For day-to-day garment inspection, you only use the General levels (G I, G II, G III). 95% of buyer inspections are G II.

Free online AQL calculator

We built a free online AQL calculator at scanerp.pro/tools that implements the full ISO 2859-1 Table II-A. Enter lot size, pick inspection level, pick AQL — it returns the sample size, accept number, and reject number instantly.

If you are building a QC app or factory ERP and want the same math in your own code, we open-sourced it as two MIT-licensed packages:

• npm: garment-aql-calculator — npm install garment-aql-calculator
• PyPI: garment-aql-calculator — pip install garment-aql-calculator

Both packages include the full Table II-A, a 3-tier garment inspection helper, and an accept/reject decision function. 34 inline tests verify every value against QIMA's published reference.

Related reading

• How we reduced our rejection rate from 8% to under 3%
• Open-source garment industry toolkit — 8 free calculators
• CMT factory ERP software for garment exports