A rejected safety data sheet is rarely empty. It is usually complete, professional-looking and wrong in one of a small number of predictable ways. This page sets out which ways, and what enforcement authorities actually found when they went looking.
What inspectors found
These are enforcement findings published by the European Chemicals Agency. Read the section on method at the foot of this page before drawing conclusions from any of them — the samples are targeted at risk and are not surveys.
Safety data sheets themselves — REF-11
The Forum’s eleventh enforcement project examined more than 2,500 safety data sheets across 28 countries, and reported in December 2024. 35% of the sheets were non-compliant.
- 67% lacked information on nanoforms where it was required.
- 48% lacked information on endocrine-disrupting properties.
- 27% carried data-quality defects in hazard identification, composition or exposure control.
- 18% were missing required exposure scenarios.
- 16% were missing information required by authorisation decisions.
Two things are worth noticing. These were predominantly EU suppliers operating inside the EU system — not importers unfamiliar with the rules. And the trend is old: comparable inspections returned around 48% two decades ago and 52% in a project from 2013. Whatever this is, it is not a new problem and it is not confined to any one region.
Sources: ECHA news release · REF-11 project report (PDF)
Imported substances and mixtures — REF-12
The twelfth project examined 2,603 import checks across 29 countries and reported in December 2025. Around one in three — 32% — of substances in imported mixtures lacked the required registration.
- 7% of substances imported on their own lacked registration.
- 16% did not comply with restrictions.
- 4 of 21 checks found a missing or expired authorisation for a substance of very high concern.
- The largest group of checked products originated in China — 1,334 cases.
ECHA reports that almost all non-compliant products detected before release for free circulation were turned back, or released only after corrective action. That is the mechanism by which a documentation defect becomes a commercial event: not a fine months later, but goods that do not move.
Sources: ECHA news release · REF-12 project report (PDF)
Products sold online — REF-8
An earlier project looked at nearly 6,000 products sold online across 29 countries. 78% were non-compliant with restrictions, and 75% of offers breached the advertising duties in Article 48 of CLP. More than 5,000 enforcement actions followed. Non-compliance was higher on marketplace platforms than in dedicated shops.
This campaign has been repeated — inspections ran in 2025 and the report is expected in 2026. Imports and online sales both remain priorities in the Forum’s programme for 2026–2027, which is worth knowing if you are planning to sell into the EU through a marketplace.
Sources: ECHA news release · REF-8 project report (PDF)
The six ways a sheet fails
Enforcement statistics say how often documents fail. They do not say what is wrong with yours. In our own reading of sheets, the defect is almost always one of these six.
It was correct when it was written
Classification changes. An ATP to CLP adds a harmonised entry, a substance moves onto the candidate list, a transport entry is amended. A sheet issued three years ago can be faultless work that no longer describes the current legal position. Nothing about the document looks wrong — the date is the only clue, and the date is easy to miss.
It contradicts itself
Section 2 carries one set of hazard statements, Section 15 implies another, and the label artwork shows a third. This happens whenever sections are written or updated separately, which is most of the time. An inspector who finds one internal contradiction reads the rest of the document differently.
The identifiers do not match
The CAS number on the sheet, the number on the invoice and the number in the customs declaration are supposed to be the same number. When they are not — a transposed digit, a hydrate confused with an anhydrous form, a salt documented as the free acid — the shipment stops while somebody works out which substance is actually in the drum.
It was written for another jurisdiction
A sheet prepared to OSHA HazCom or to a Chinese national standard can be excellent and still not satisfy REACH Annex II. The section count differs, the required content differs, and translating the text does not fix a structural mismatch. This is the most common problem we see on first-time exports into the EU.
Section 14 is missing what it needs
Transport information is the section that stops shipments, because it is the one a carrier reads before accepting the goods. UN number, proper shipping name, packing group and the mode-specific entries have to be there and have to agree with the classification in Section 2.
A value was estimated and presented as measured
This is the failure that does real damage, because it survives every casual review. A property that was never measured for this substance appears as a clean number with no qualification. It reads as authoritative until somebody checks the source, and then the whole document is in question.
It is also the reason we mark fields unavailable rather than filling them. A gap is an inconvenience. A confident wrong number is a liability.
How these findings were arrived at
This section stays on the page in full. It is the part a careful reader checks first, and the part that disappears first when a page is shortened for effect.
Enforcement samples are targeted at risk, not drawn at random. Market-surveillance projects select what to inspect using intelligence about where problems are expected: product categories with a history of non-compliance, importers previously found at fault, substances under recent restriction. A sample built that way finds a higher rate of defects than exists in the population, by design — that is what makes it efficient enforcement and what makes it unsuitable as a population estimate.
Figures from different projects are not comparable with each other, and must not be combined. Scope, definitions of non-compliance, the member states taking part and the inspection method all change between projects. A percentage from REF-11 placed beside a percentage from REF-8 does not produce a trend, an average or any aggregate indicator. We do not compute one and neither should anyone quoting this page.
The predominance of products from China in REF-12 reflects trade volume and the targeting of checks — not a measured failure rate per country. That 1,334 is a count of what was inspected, not a denominator anybody has published. Reading it as a national quality score is the single easiest mistake to make with this dataset.
What such findings do support. That specific, named defects were found repeatedly in documents that had already been issued and were already in commercial use, and that goods were turned back as a result. That is the claim this page makes.
What we do not claim
We do not state a share of documents in the market that would fail. We do not present enforcement findings as a survey. We do not convert them into a probability that your sheet is defective.
And specifically: there is no official figure stating that 90% of Chinese safety data sheets are rejected. We do not publish one, and you should treat any supplier who quotes it as having invented it. It is the number this industry repeats most often and the one with the least behind it.
If you want to know about your sheet, the audit reads your sheet.
Three sheets, no charge
Find out which one is yours
Send up to three safety data sheets you already hold. Within one working day you get a written report naming what is missing, what contradicts itself and what is out of date, with the source for every point.
Questions people ask before they order
Where does the figure of 35% come from?
From ECHA enforcement reporting across 28 participating countries. The sample is risk-targeted rather than random, which means it is not a clean estimate of the whole market — inspectors look where they expect to find something. That caveat belongs with the number every time it is quoted.
Does that mean a third of chemical companies are careless?
No, and the failure modes suggest otherwise. Most defects are structural rather than negligent: a template written before the last CLP adaptation, a section copied between products, a value that was correct when it was first recorded. Nobody was told when the entry changed.
What is the most common defect?
A classification that no longer matches the harmonised entry, followed by sections that contradict each other inside the same document. Both are invisible to the person who wrote the sheet and obvious to somebody reading it against the register.
How would I know whether my own sheets have this problem?
Send three of them. One sheet shows an incident; three show whether the defect lives in the template you generate everything from. It costs nothing and takes one working day.