Don’t take our word for it. Here’s every fact on this site, with its evidence grade, its caveats, and a link to the source, so you can check us.
We start you on the strong ones: findings that have been replicated or come from one large, well-controlled study. Moderate and weak facts are here too, labeled for what they are. Use “Link to this fact” to share a single card.
The industry / Lab study (cells or chemistry)
The fragrance industry's own data show its main explanation for how fragrance chemicals cause allergy doesn't cover most of them: 34 of 62 ingredients (55%) sensitize despite having no predicted protein-binding alert.
Moderate evidence.
Caveats
Structural-alert software limitation; metabolic activation explains some. Roberts & Api 2018 (different 57-chemical fragrance dataset, no direct overlap confirmed with this claim's 62-chemical set) adds two further indirect-activation routes the alert software here evidently does not capture — slow autoxidation of tertiary allylic/benzylic C-H compounds to allergenic hydroperoxides/epoxides, and a proposed bio-sulfation route for benzylic alcohols — which could explain part of the unexplained 55% for chemicals with those structural motifs. The same paper separately proposes that some LLNA-positive, alert-free fragrance chemicals (hexyl/benzyl salicylate) may be LLNA false positives rather than true sensitizers at all, which would pull in the opposite direction (shrinking, not explaining, the true no-mechanism fraction) if any such chemicals are present in this claim's 62-ingredient set — not checked here, so the net effect on the 55% figure is unresolved and the grade is left unchanged.
Source: Lee 2024, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
Those models underestimated allergy potency by more than fivefold for 18% of the 188 chemicals tested, and more than tenfold for 11%.
Moderate evidence.
Caveats
Prediction error, not a safety-limit error; worst outliers are non-fragrance chemicals. Natsch 2023 (same author, human-anchored retraining) shows this outlier structure is not simply an artifact of fitting to mouse data: only 6/139 chemicals (4.3%) shift by more than 3.3-fold when the same models are retrained against human-anchored potency values instead of LLNA.
Source: Natsch & Gerberick 2022, ALTEX Industry-funded Link to this fact
The industry / Review of other studies
The industry panel judged each salicylate's safety one chemical at a time, although the same report says all 17 are broken down in the body to the same substance, salicylic acid.
Moderate evidence.
Caveats
The combined-exposure figure (about 1.05 mg/kg/day, group margin roughly 48-69 at the panel's own 100%-absorption default) is a evidence base calculation on the paper's Table 1, not a published number, and sums high-end maxima.
Source: RIFM Expert Panel / Belsito 2007, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Measured in people or real products
In a French fragrance-and-flavor factory, the company's own measurements found the butter-flavor chemical diacetyl (linked to 'popcorn lung' in US factory workers) in workers' breathing air at 0.05–9 ppm during short tasks; 29 of 31 samples were above the 15-minute limit of 0.1 ppm recommended by the EU's scientific committee on occupational limits.
Moderate evidence.
Caveats
Limit comparison is our (paper makes none); samplers on clothing, respirator use per task not reported; flavor chemical, factory workers, not consumers; industry-funded (V. Mane Fils).
Source: Angelini 2016, PLoS ONE Industry-funded Link to this fact
The industry / Measured in people or real products
The standard EU (REACH) screening tool for estimating workers' chemical exposure predicted lower levels than were actually measured in 61% of 431 measurements in a fragrance-and-flavor factory, sometimes by more than 100-fold.
Moderate evidence: One plant; evidence base count from the paper's S4 table (paper: 37% of tool predictions were above the measurement); short task samples.
Source: Angelini 2016, PLoS ONE Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
When the same lab-dish tests that replace animal testing are checked against real human allergy-test panels instead of mouse data, they explain only about 32-45% of how strongly a chemical causes allergy in people, and the mouse test itself does barely better, explaining only 43%.
Moderate evidence: The strictly-human comparison dataset (n=62) is small and the author flags it as biased/non-representative; a larger human dataset might do better.
Source: Natsch 2023, ALTEX Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
Whether a chemical is irritating to cells in a dish, not just whether it reacts with proteins, drives a lot of the mouse allergy test's readout in ways that don't reflect real human allergy.
Moderate evidence: Statistical inference from regression coefficients across three chemical subsets, not a direct mechanistic experiment.
Source: Natsch 2023, ALTEX Industry-funded Link to this fact
The industry / Measured in people or real products
In RIFM's own comparison of 57 fragrance chemicals, the mouse test used to set fragrance skin-allergy safety limits agreed with real human sensitization thresholds within about 3-fold for 76% of the chemicals tested, but was off by up to nearly 800-fold for others, including hexyl and benzyl salicylate.
Moderate evidence.
Caveats
Single paper/author group (RIFM staff plus an industry consultant), though a large (57-chemical), systematically protocolled primary dataset; the authors themselves call the outlier groups 'an alert for consideration, rather than… a general rule' given only 7 materials populate each. Corroborates, from a different metric and era, the LLNA-human accuracy gap C230 documents for the newer non-animal replacement tests.
Source: Api, Basketter & Lalko 2014, Cutaneous and Ocular Toxicology Industry-funded Link to this fact
The industry / Measured in people or real products
A musk fragrance chemical called ATTN (Versalide) disappeared entirely from human tissue in this 1996 study because industry had already stopped making it in 1980, after animal testing showed it damaged nerves – a real-world example of a musk being phased out after a toxicity finding.
Moderate evidence: Absence in 19 samples is consistent with, but does not by itself prove, the discontinued-production explanation the authors give; based on a small single-country sample.
The industry / Animal study
When RIFM's own scientists restrict the comparison to fragrance chemicals without known problematic chemistry, the mouse test used to set fragrance safety limits correlates strongly with real human sensitization thresholds (R-squared = 0.78 for 35 chemicals).
Moderate evidence.
Caveats
This correlation excludes, by design, the 14/57 chemicals already known (from the same underlying RIFM dataset) to disagree badly between mouse and human data, some by up to ~790-fold (Api 2014). Only 2 of those 14 exclusions could have been identified before seeing the human data; the other 12 were fit to the very outliers they exclude.
Source: Roberts & Api 2018, Regulatory Toxicology and Pharmacology Industry-funded Link to this fact
The industry / Measured in people or real products
This foundational human PGME irritation study has a direct Dow Chemical link stronger than later PGME papers in the wiki: one author was a Dow Chemical employee and the test material was Dow-supplied ('Dowtherm 209'), yet the study documents substantial irritation at concentrations relevant to Dow's own radiator-coolant product rather than minimizing it.
Moderate evidence: A single industry-linked study reporting an adverse finding is still one study; the funding link is recorded for transparency, not as proof the finding is unbiased in general.
Source: Stewart 1970 (Arch Environ Health), Archives of Environmental Health: An International Journal Industry-funded Link to this fact
The industry / Other evidence
In a random sample of 100 high-production-volume industrial chemicals, 47 were also listed on EPA's Toxics Release Inventory (meaning they are emitted into air, land, or water in quantities over 10,000 pounds a year), and toxicity data were publicly available for less than half of those 47.
Moderate evidence: A single external audit, not independently replicated; concerns industrial chemicals broadly, not a named fragrance or consumer-product ingredient.
Source: Goldman 1998, Environmental Health Perspectives Link to this fact
The industry / Animal study
The paper's proposed inhalation systemic TTC ranges from 16.4 μg/kg/day for the least-toxic Cramer Class 1 chemicals down to 2.8 μg/kg/day for the more toxicity-presumptive Class 3 chemicals (roughly a 6-fold spread), while its separate local respiratory-tract TTC ranges from 2.1 to 0.73 μg/g lung tissue/day (about a 3-fold spread) — equivalent to 980/170 μg/day and 1400/470 μg/day respectively for a 60 kg adult.
Moderate evidence.
Caveats
Derived from the 5th percentile of NOAEC/NOAEL across 92 rat studies plus a 25-fold uncertainty factor; a conservative statistical screening threshold, not a measured no-harm level, and the paper treats local and systemic thresholds as two separate comparisons that later citations sometimes conflate.
Source: Carthew, Clapp & Gutsell 2009, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Animal study
The dataset behind these inhalation TTCs excluded genotoxic carcinogens, in vivo mutagens, heavy metals, dioxins/PCBs, organophosphates, and polymers — chemical classes the authors say would already be disqualified from consumer aerosol use on other safety grounds — before applying a default uncertainty (safety) factor of 25 to the 5th-percentile NOAEC/NOAEL of the remaining 92 rat studies.
Moderate evidence.
Caveats
Exclusion criteria and the 25-fold uncertainty factor are the authors' own choices, following precedent from the oral/dietary TTC literature; not independently audited by a regulator.
Source: Carthew, Clapp & Gutsell 2009, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Animal study
Only 5 of the 92 chemicals in the database had local (respiratory-tract) NOAECs below the study's own 5th-percentile benchmark — sulfuric acid, trimellitic acid, glutaraldehyde, hydrogen peroxide, and 1,2-dichloro-4-nitrobenzene — and all were strong acids, bases, or reactive irritants that the authors say would already be excluded from consumer aerosol use on physicochemical grounds alone.
Moderate evidence.
Caveats
A post-hoc observation about which chemicals happened to fall below the benchmark in this specific 92-study set; not a claim that physicochemical screening alone catches every locally toxic inhalable chemical.
Source: Carthew, Clapp & Gutsell 2009, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Review of other studies
When RIFM's own scientists switched from mainly human patch-test data (their 2017 categorization) to a full weight-of-evidence approach that also folds in animal, computer-model, and lab-dish data, the assigned allergy-potency category changed for 29% of the same 82 fragrance chemicals — moving to a weaker category for 20.5% of them and a stronger one for 8.5%.
Moderate evidence.
Caveats
Same authors and institution comparing their own two categorization schemes; independent validation of either scheme against real-world human sensitization rates is not addressed in this paper.
Source: Na et al. 2022, Dermatitis Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
In lab dish tests, three human smell receptors that detect the sulfur gases methanethiol and hydrogen sulfide (OR2T1, OR2T6, OR2T11) only responded strongly when copper was added to the mix, and a copper-grabbing chemical shut the response back down, indicating these malodor receptors sense sulfur gases by using copper ions to grab onto them.
Moderate evidence.
Caveats
Heterologous cell-culture assay only, at concentrations far higher than natural environmental levels, per the authors' own discussion; not directly observed in olfactory neurons or in vivo.
Source: Fukutani et al. 2023, Current Biology Industry-funded Link to this fact
The industry / Government agency
Benzophenone is widely used as a UV-curing photoinitiator for inks and varnishes, a fragrance fixative/enhancer and flavoring agent, a UV blocker in plastic packaging, and an additive in coatings, adhesives, laundry and household cleaning products; the US EPA classified it a high-production-volume chemical in 2003, with US annual production exceeding 1 million pounds (453,000 kg).
Moderate evidence: Production/use figures are the Working Group's synthesis of industry and HSDB/NTP secondary sources, not independently verified primary measurement.
Source: IARC 2013 Link to this fact
The industry / Lab study (cells or chemistry)
Cosmetics Europe (the European cosmetics industry trade association) funded a standardized, OECD-compliant lab study measuring 24-hour skin penetration of 56 cosmetic-relevant chemicals, authored by scientists employed by L'Oreal, Unilever, Pierre Fabre, Procter & Gamble, BASF, Coty, and Beiersdorf, with the explicit goal of replacing a flat 50% default assumption about how much of an applied chemical passes through skin with chemical-specific measured data.
Moderate evidence.
Caveats
Industry-funded and industry-authored throughout; the paper states 'the authors state no conflict of interest,' which addresses personal financial conflicts, not the industry funding and employment of the entire author list.
Source: Hewitt 2020, Journal of Applied Toxicology Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
For the 7 chemicals in this dataset with published human in vivo comparison data (hydrocortisone, testosterone, ibuprofen, benzoic acid, caffeine, DNCB, and nitrobenzene — none of them fragrance ingredients), measured skin penetration ranged from about 3% to 63% of the applied dose, a more than 20-fold spread that straddles the industry's flat 50% dermal-absorption default from well below to just above it.
Moderate evidence.
Caveats
Only 7 of the 56 chemicals had in vivo comparison data available; none of the 7 is a fragrance ingredient; single industry-funded lab, though methodology follows OECD test guideline 428.
Source: Hewitt 2020, Journal of Applied Toxicology Industry-funded Link to this fact
The industry / Lab study (cells or chemistry)
In this standardized dataset, several volatile test chemicals lost most of their applied dose to evaporation from the skin surface rather than to skin absorption — naphthalene's mass balance was only 27% of the applied dose and 4-tolunitrile's only 20% — but the study measures only dermal delivery (skin plus receptor fluid) and never quantifies or discusses the airborne exposure that this evaporation implies for someone wearing a product containing these chemicals.
Moderate evidence.
Caveats
This is a methodological gap in the study's design (dermal-only endpoints), not a measured inhalation finding; extends our C178 entry on the fragrance/cosmetics industry's recurring lack of an inhalation branch in its safety-assessment research.
Source: Hewitt 2020, Journal of Applied Toxicology Industry-funded Link to this fact
The industry / Other evidence
A fragrance fixative left on skin for hours in a lotion or perfume penetrates far more than the same chemical in a soap that is quickly rinsed off: an FDA lab's modeled estimate for musk xylol found roughly 50-60 times more absorbed mass from a 12-hour lotion application than from a 4-minute soap wash, at realistic in-use concentrations and body-surface areas.
Moderate evidence: Modeled, not directly measured for this scenario; the Kp was derived from more-permeable hairless guinea pig skin, which the authors themselves say overestimates human absorption.
Source: Hood 1996, Food and Chemical Toxicology Link to this fact
The industry / Other evidence
The flavor trade association says workplace exposure to flavor chemicals, at 'relatively high concentrations' by dermal and inhalation routes, is not evaluated by its Expert Panel and falls to OSHA; it names the butter-flavor chemical diacetyl as a 'potential respiratory hazard' in flavor and popcorn manufacturing while it remains GRAS for its food flavor use.
Moderate evidence.
Caveats
The paper reports no workplace measurements or health data itself. Written by two FEMA employees and supported by FEMA (the flavor industry trade association); it is the industry's own account of its own program.
Source: Hallagan 2009, Food and Chemical Toxicology Industry-funded Link to this fact
The industry / Other evidence
Skin Deep's own FAQ says its product hazard ratings do not account for how much of a chemical you are exposed to or how sensitive you are, so they describe ingredient hazard, not your actual risk.
Moderate evidence: EWG's own description of its method; no independent audit. The score is still a label-based proxy.
Source: EWG, Skin Deep FAQ (web page, accessed 2026-10-16), EWG web page (Skin Deep FAQ) Link to this fact
The industry / Other evidence
The 2006 ADBAC reregistration database was supported by a registrant consortium (Lonza, Mason Chemical, Stepan), with EPA reviewing the data it submitted.
Moderate evidence: EPA reviewed the studies; the point is that the underlying studies are registrant-submitted, not independent.
Source: EPA RED 2006 Link to this fact
Information, not medical advice. See also: myths we won’t tell you.