Safety · 8
The EU names 26 fragrance allergens; the database holds 23 of them, and two of the three missing were later banned
· 23 min read
Every so often a headline says a perfume was found to contain an EU-banned sensitiser. So I took the list of fragrance allergens that Annex III of the EU cosmetics regulation requires to be named on the label, and checked all 26 against this database one at a time. 23 are there, with 1,774 suppliers between them and geraniol the largest at 198. The three missing are HICC (Lyral), BMHCA (Lilial) and methyl 2-octynoate, and the first two are precisely the ones the EU later banned. A catalogue never marks a material as banned; it simply stops appearing, and absence is not a warning. Two more things: of the 23, only 13 carry skin sensitisation H317 in the GHS field, with linalool, coumarin and benzyl alcohol among those that do not, because a labelling list and a hazard classification are two lists with different purposes. And only 359 of the site's 42,225 records carry a GHS field at all, under 1%. Two papers: European data on 124,472 patients shows HICC allergy falling year on year before the ban took effect, and a 2024 systematic review puts fragrance mix I and II sensitisation at 6.81% and 3.64%.
About a 5 minute read.
Too long, didn't read
- Annex III of the EU cosmetics regulation requires a set of 26 fragrance allergens to be named on the ingredient list once they pass a threshold
- 23 of the 26 are in this database, with 1,774 suppliers between them; the largest is geraniol at 198
- The three that are not: HICC (Lyral), BMHCA (Lilial) and methyl 2-octynoate
- The first two are exactly the ones the EU banned. A catalogue does not mark a ban, it just stops listing
- Of the 23, only 13 carry H317 skin sensitisation in the GHS field
- Of all 42,225 records on the site, only 359 have a GHS field at all, under 1%
Labelling is not banning
Two things get mixed together, so separate them first.
Labelling means that once a material passes 0.001% in a leave-on product or 0.01% in a rinse-off one, the ingredient list can no longer hide it inside "parfum" and has to name it. The purpose is to let somebody who already knows what they react to work out which bottle on the shelf to avoid. It is disclosure, not prohibition.
Banning is a separate process. HICC has not been permitted in cosmetics since August 2021, and BMHCA, better known as Lilial, since March 2022, the latter because it was classified as toxic for reproduction.
Headlines that read "found to contain an EU-banned sensitiser" glue the two concepts together. Almost all 26 on the list remain legal today, and they are in the bottles on your bench.
What the 26 look like in the database
Checked one at a time. Longevity is hours at 100%; the H317 column asks whether the GHS classification names skin sensitisation.
| Labelled as | Name in the database | Suppliers | Longevity | H317 |
|---|---|---|---|---|
| geraniol | geraniol | 198 | 60 | yes |
| citral | citral | 135 | 12 | yes |
| linalool | linalool | 135 | 12 | no |
| benzyl alcohol | benzyl alcohol | 125 | 35 | no |
| citronellol | citronellol | 119 | 56 | yes |
| cinnamal | cinnamaldehyde | 105 | 212 | yes |
| eugenol | eugenol | 97 | 52 | yes |
| benzyl benzoate | benzyl benzoate | 84 | 322 | no |
| limonene | dextro-limonene | 82 | 4 | yes |
| cinnamyl alcohol | cinnamyl alcohol | 79 | 371 | yes |
| coumarin | coumarin | 74 | 364 | no |
| anise alcohol | para-anisyl alcohol | 59 | 82 | no |
| benzyl salicylate | benzyl salicylate | 58 | 384 | yes |
| hexyl cinnamal | alpha-hexyl cinnamaldehyde | 56 | 400 | yes |
| isoeugenol | isoeugenol | 56 | 400 | yes |
| amyl cinnamal | alpha-amyl cinnamaldehyde | 53 | 256 | yes |
| hydroxycitronellal | hydroxycitronellal | 53 | 218 | yes |
| alpha-isomethyl ionone | alpha-isomethyl ionone (50% min.) | 47 | 124 | yes |
| benzyl cinnamate | benzyl cinnamate | 45 | 243 | no |
| oakmoss | oakmoss absolute | 42 | 400 | no |
| farnesol | farnesol | 31 | 400 | no |
| treemoss | treemoss absolute | 29 | 400 | field empty |
| amylcinnamyl alcohol | alpha-amyl cinnamyl alcohol | 12 | 320 | field empty |
| HICC (Lyral) | not found | — | — | — |
| BMHCA (Lilial) | not found | — | — | — |
| methyl 2-octynoate | not found | — | — | — |
Those 23 have 1,774 suppliers between them. This is not an obscure list; it is most of a starter palette. Geraniol, linalool, coumarin and eugenol are all on it.
The three that are missing
HICC and BMHCA are absent not because we failed to load them, but because they are no longer in the upstream catalogue. Both were banned in the EU and then disappeared from commercial listings.
That has an unhelpful shape for anyone using the data. A catalogue does not tell you a material has been banned; it simply stops appearing. No red text, no note, no "withdrawn" field. You could buy it last time and cannot find it this time, and the database does not explain what happened in between. The Lilial article is the single-material version of this trap; this is the systematic version, after running the whole list.
Absence is not a warning. To judge whether a material may be used today, read the current regulation, not the catalogue. The reverse holds too: presence in a catalogue does not make something legal in your market.
The third one, methyl 2-octynoate, is not banned. Our source data simply does not carry it. Two of the three gaps come from bans and one from coverage, and the two kinds of absence look identical.
124,472 patients: HICC allergy was already falling before the ban
In 2021 Ahlström and colleagues at Gentofte Hospital in Copenhagen and at Erlangen published a time-trend analysis in Contact Dermatitis (PMID 33453125).
Their background sentence is worth copying first: HICC has been the most reported fragrance chemical for two decades, and would be prohibited in cosmetic products from August 2021.
The scale is 124,472 dermatitis patients, all patch tested with HICC 5% in petrolatum in the baseline series, from the European Surveillance System on Contact Allergy (ESSCA) between 2009 and 2018 and from Herlev-Gentofte Hospital between 2009 and 2019.
The results: contact allergy to HICC was found in 1.98% of 9,865 patients in Gentofte and 1.62% of 114,607 in ESSCA. The prevalence decreased annually, by 0.156 percentage points in Gentofte (P = .001) and 0.051 percentage points in ESSCA (P = .0002).
The authors' conclusion is restrained: this is the first study to demonstrate a significant decline in HICC allergy in European dermatitis patients, most likely attributed to the upcoming European ban. Formulators moved before the regulation took effect, and the sensitisation rate in the population followed.
A systematic review of 84 studies: 6.81% and 3.64%
The second paper is a 2024 systematic review by much the same group (Botvid et al., PMID 38945918). Following a protocol registered in advance on PROSPERO, they screened 4,134 publications of patch test results in European dermatitis patients from 1981 to 2022 and analysed 84 eligible original articles.
The results: sensitisation to fragrance mix I was 6.81% (95% CI 6.37–7.28) and fragrance mix II 3.64% (3.3–4.01). The geography differs, with FM I most prevalent in Central and Eastern Europe and FM II in Western Europe.
Both numbers need placing carefully. They come from dermatitis patients in dermatology clinics, not the general population, so the denominator is skewed high to begin with. This 2024 paper also carries a 2026 erratum, which should be read alongside it.
The GHS field and the labelling list are different lists
Back to the right-hand column of that table. Of the 23 labelled allergens present, only 13 carry H317 skin sensitisation in the database's GHS field. Linalool does not, coumarin does not, benzyl alcohol does not, oakmoss absolute does not.
That is not a contradiction; the two lists were built for different purposes. The labelling list asks whether a substance is epidemiologically worth showing to consumers. GHS classification asks which hazard statement a substance earns on its test data. The evidence bar and the decision process differ, so agreement is not required.
The coverage is the part to remember. Of 42,225 materials on the site, 359 carry a GHS field, under 1%, and 63 of those name H317. So for the overwhelming majority of records, the correct reading of "no sensitisation in the GHS field" is that field is empty, not "safe". Drawing a safety conclusion from a field with under 1% coverage is the trap are naturals gentler already walked into.
What to do with this in practice
- If you hold any of the 23, treat it as something that will need naming. For your own use it makes no difference; the moment you give or sell it to someone, the ingredient list obligation starts.
- Check the current regulation, not the catalogue. The EU list was revised in 2023 and now covers more than these 26, and rules differ by market. Work from the official text.
- The concentration on a blotter is not the concentration on skin. The 0.001% threshold applies to a finished product, which is not the same coordinate as the percentages in your dilution work.
- Avoid them if you want to, but do not read the list as a danger ranking. These 26 were selected on epidemiology to be made visible, not sorted by toxicity.
→ Search by odour: all 23 are in the database with full physical data and suppliers, starting with the largest on the list, geraniol.
What this article does not establish
- The match against the 26 was done by name. Database naming and regulatory naming do not align perfectly (cinnamal against cinnamaldehyde, limonene against dextro-limonene). I checked each one, but name matching can miss.
- For HICC and BMHCA I inferred that they left the catalogue after being banned. I have no evidence about the source data's inclusion decisions, only the result that they are not there.
- methyl 2-octynoate being absent does not mean it is banned; it is simply not in our source.
- 1.98% and 1.62% are proportions among dermatitis patients in clinics, not prevalence in the general population. The same applies to 6.81% and 3.64%, and that paper carries an erratum.
- Regulatory detail changes. This was written in October 2026; the two ban dates come from the literature and published rules. Work from the current official text.