A pre-shipment inspection dimension check fails orders that are, physically, perfectly fine. The goods are well built. The finish is clean. Then the inspector puts a tape on a table top, writes down 1198 mm where your catalog says 1200 mm, and the report comes back "FAIL — dimension nonconformity." Your container doesn't load, the buyer holds the balance, and you're arguing about two millimetres you never promised to hold in the first place.
The pattern is almost never a factory problem. It's a documentation problem: the inspector needs a number to measure against, and whatever you published becomes that number — tolerance or no tolerance. What happens next depends entirely on which of the following situations you're in.
How a pre-shipment inspection dimension check actually works
Third-party inspections run on sampling standards, and there are two different families. Getting them confused is where a lot of the arguing starts.
| Standard | Inspection type | What the inspector records | Typical use |
|---|---|---|---|
| ISO 2859-1 | By attributes | Conforming / nonconforming — a count | Almost all consumer-goods PSI |
| ISO 3951-1 | By variables | The actual measured value on a continuous scale | Critical dimensions, statistical process control |
Here's the part that catches people: a dimension is a measured value, but a normal inspection converts it into a pass/fail attribute. The inspector doesn't report "1198 mm" as data to be analysed. They compare 1198 against your stated spec, decide "nonconforming," and add one to a defect count. That count runs against an Acceptance Quality Limit (AQL) — the maximum proportion of nonconforming units the sampling plan will tolerate before the lot is rejected.
Which means the entire outcome hinges on one thing: what you published as the spec, and whether you published a tolerance with it. No tolerance means the comparison is "1198 = 1200?" and the answer is no.
That's the sentence worth pinning above the desk: an inspector doesn't judge whether your product is good — they judge whether it matches the number you published.
Scenario A: Your spec sheet gives dimensions with no tolerance
If this is you — your catalog and PO say "1200 × 600 × 750 mm" and nothing else — the inspector has no band to work with. What they do next is not standardised, and that's the danger. Some apply the buyer's default tolerance from a QC checklist you've never seen. Some apply their own house default. Some measure, note the deviation, and let the buyer decide after the fact.
Do this instead: publish a tolerance next to every dimension a buyer could measure. Not a blanket footnote — a per-dimension band, tightest where fit matters and loose where it doesn't.
| Dimension type | Realistic band to publish | Why |
|---|---|---|
| Overall width / depth / height | ±2–5 mm on furniture-scale parts | Wood moves; nobody assembles to the outside face |
| Interface / mating dimension | as tight as you truly hold it | This is what actually fails to fit |
| Fabric, foam, upholstered parts | wider, and say so | Compressible material genuinely varies |
| Packed carton | ±5–10 mm | Board thickness and stuffing vary by run |
Expected result: 1198 against a published "1200 ±3 mm" is conforming. Same goods, same tape measure, opposite report. If you've never set these bands, work through how to label ± specs in product dimension tolerance before your next inspection is booked, not after.
Scenario B: Your contract still cites ISO 2859-1:1999
If this is you, your paperwork is now out of date. ISO 2859-1 was revised: ISO 2859-1:2026 is the third edition, published 2026-01, and it replaces the 1999 edition along with its amendments. The new edition introduces skip-lot sampling procedures and updated guidance on applying sampling strategies.
Most supplier contracts, QC manuals, and inspection booking forms still name the 1999 edition, because it stood unchanged for twenty-six years. Nothing detonates on its own — but a contract citing a superseded edition is exactly the loose thread that gets pulled during a dispute.
Do this instead: check what edition your inspection clause names. Agree it explicitly with the buyer along with the inspection level and the AQL you're both accepting, and write the agreed values into the PO rather than leaving them to a booking form.
Expected result: you and the buyer are inspecting against the same rulebook. It also puts you ahead of most of your competition, who are quoting a withdrawn edition without noticing.
Scenario C: The golden sample was approved, then production drifted
If this is you — sample approved months ago, mass production runs 3 mm over — you're in the most expensive version of this problem, because the approved sample itself is now evidence against you.
Under international sales law this is explicit, and it surprises suppliers every time. The UN Convention on Contracts for the International Sale of Goods states at Article 35(2)(c) that goods do not conform unless they "possess the qualities of goods which the seller has held out to the buyer as a sample or model." The sample you sent is the specification, whether or not anyone wrote its dimensions down.
Do this instead: measure the golden sample and record its actual dimensions on the approval document — the real measured values, not the catalog's nominal ones. Both sides sign that sheet. That converts "it should look like the sample" into a number with a band around it.
Expected result: production is checked against a documented measurement instead of against a physical object sitting in someone's showroom being remembered generously.
Scenario D: The dimensions are right but the inspector still wrote them up
If this is you, you're probably being measured on a different state of the product than the one you published. The classic set:
- You published the assembled size; the inspector measured the flat-packed carton.
- You published the product; they measured product + feet + handle projection.
- You published inside capacity; they measured outside.
- Your drawing showed the seat height unloaded; they measured it compressed.
None of these are the factory's fault, and all of them read as "dimension nonconformity" in a report.
Do this instead: label the state, not just the number. "1200 mm (assembled, excl. handle)" removes the argument before it starts. If your product ships in one state and is used in another, both belong on the sheet — the reasoning in nominal vs actual dimensions applies directly here: the name of a size and the measurement of a size are different things, and inspectors measure the second one.
Expected result: the inspector measures what you meant, because you told them which measurement you meant.
Decision matrix
Every pre-shipment inspection dimension check reduces to one of these five situations:
| Your situation | The real risk | The move |
|---|---|---|
| Dimensions published without tolerance | Inspector applies someone else's default | Publish per-dimension ± bands |
| Contract cites ISO 2859-1:1999 | Superseded edition; disputes get messy | Re-cite ISO 2859-1:2026, agree AQL + level in the PO |
| Sample approved, no measurements recorded | Sample becomes the spec under CISG Art. 35(2)(c) | Measure and sign off the golden sample |
| Assembled vs packed vs projected size confusion | "Nonconformity" that isn't one | Label the state alongside every number |
| Critical fit dimension treated like any other | One bad unit fails the whole lot | Flag it as critical, tighten the band, say so |
What a failed inspection actually costs
Suppliers underprice this because the inspection fee is small and visible while everything else is large and invisible. A failed lot typically means re-inspection fees, a missed vessel and rebooked freight, storage while it sits, and a buyer who now discounts, delays the balance, or walks. For anything bulky — furniture, building materials, industrial units — the freight and storage lines dwarf the goods themselves. Running the actual per-order number through a return cost calculator is a five-minute exercise that reliably ends the debate about whether tolerance documentation is worth doing.
Next steps
Pick whichever of these you don't already have:
- Add tolerances to your top 20 SKUs. Not all of them — the twenty that ship most. Per-dimension bands, honest ones you actually hold.
- Re-read your inspection clause. Which edition, which inspection level, which AQL, who books it, who pays for re-inspection. Agree it in the PO.
- Measure your golden samples and sign the sheet. The sample is already your specification under CISG Article 35(2)(c) — you may as well control what it says.
- Put the numbers where the buyer and the inspector both look: on the image. A spec sheet in an email attachment gets lost; the dimension printed on the product photo travels with the listing, the quote, and the inspection brief. If you're producing these by hand, dimension and spec annotation software will hold the label to the product's measured edge and let you re-export the same diagram at each destination's size — which matters here specifically because a dimension diagram is only useful if the number on it is the number you actually hold. AI image tools are the wrong instrument for this job: they restyle a photo, and any measurement they produce is a plausible guess, not your tolerance.
- Or do none of it and keep arguing over two millimetres. It's a legitimate choice, right up until the vessel sails without your container.
An inspection is not a quality opinion. It's a comparison against a document — so the document is the part you control.
FAQ
What is a pre-shipment inspection dimension check?
It's the part of a third-party pre-shipment inspection where an inspector measures sampled units and compares each dimension against the specification in your purchase order, drawing, or catalog. If a measurement falls outside the stated spec, the unit is recorded as nonconforming and counted against the agreed AQL. If the lot's defect count exceeds the acceptance number, the lot is rejected — regardless of whether the product is otherwise well made.
Why did my order fail inspection when the product is fine?
Almost always because the published dimension had no tolerance. "1200 mm" with nothing after it means 1198 mm is a deviation, and the inspector has no agreed band that makes it acceptable. Publishing "1200 ±3 mm" turns the same measurement into a pass. The goods didn't change; the document did.
Which ISO standard applies to inspection dimensions — 2859 or 3951?
Both can, and they're different. ISO 2859-1 is sampling by attributes: each unit is judged conforming or nonconforming and the defects are counted. ISO 3951-1 is sampling by variables: the actual measured values are used statistically, and it assumes the characteristic is measurable on a continuous scale with defined specification limits. In practice most pre-shipment inspections apply ISO 2859-1 and convert each dimension to a pass/fail against your tolerance.
Has the AQL sampling standard changed recently?
Yes. ISO 2859-1:2026 was published in January 2026 as the third edition. It replaces ISO 2859-1:1999 and its amendments, adds skip-lot sampling procedures, and updates guidance on applying sampling strategies. Contracts and QC manuals that still cite the 1999 edition are referencing a superseded document.
Does an approved sample count as a specification?
Yes, under the CISG. Article 35(2)(c) provides that goods do not conform to the contract unless they possess the qualities of goods the seller "has held out to the buyer as a sample or model." A golden sample is a specification whether or not its dimensions were ever written down — which is why measuring it and recording the values on the signed approval sheet protects you rather than exposing you.
Sources & References
- ISO 2859-1:2026 — Sampling procedures for inspection by attributes, Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot inspection — third edition, published 2026-01; introduces skip-lot sampling and replaces the 1999 edition and its amendments
- ISO 2859-1:1999 — the superseded second edition — the version still cited in many supplier contracts and QC manuals
- ISO 3951-1:2022 — Sampling procedures for inspection by variables, Part 1 — single sampling plans indexed by AQL for a single quality characteristic measured on a continuous scale
- UNCITRAL — United Nations Convention on Contracts for the International Sale of Goods (Vienna, 1980) — Article 35 on conformity of the goods, including 35(2)(c) on samples and models