Scared of Fluoride? US Tap Water Levels Show No Brain Harm
Peer-reviewed evidence does not demonstrate cognitive harm from fluoride at the 0.7 mg/L used in US water fluoridation; confirmed harm is limited to exposures generally above 1.5 mg/L, while claims of low-level harm rest on urinary biomarkers and extrapolation from higher-dose studies.
Does peer-reviewed evidence demonstrate cognitive harm from fluoride at the concentrations used in US municipal water fluoridation (0.7 mg/L), or do claims of harm rely on extrapolation from studies conducted at higher exposure levels?
- 1The National Toxicology Program found with moderate confidence that fluoride above 1.5 mg/L is linked to lower IQ in children, but stated the data were insufficient to judge the 0.7 mg/L level used in US water supplies.
- 2The three largest studies conducted at actual fluoridation levels—a 26,820-student US analysis, a 357-person Australian cohort, and a 2,001-woman Canadian pregnancy cohort—found no harm and, in the US case, slightly better cognition.
- 3The 2025 JAMA Pediatrics meta-analysis of 74 studies found a significant inverse fluoride-IQ association overall, yet when fluoride was measured in drinking water the negative association was significant only above 2 mg/L; only urinary measures showed effects at lower levels.
- 4A 2023 dose-response meta-analysis found the inverse association was strongest in high-bias studies and absent in the single low-bias study, and its urinary estimate near the equivalent of 0.7 mg/L water was not statistically significant.
- 5A federal court found unreasonable risk under chemical-safety law in November 2024, EPA appealed and opened a review in April 2025, and FDA warned against fluoride supplements for the youngest children in October 2025.
Peer-reviewed evidence does not demonstrate cognitive harm from fluoride at the 0.7 mg/L concentration used in US water fluoridation. The confirmed harm is limited to exposures generally above 1.5 mg/L—a level more than double what most Americans drink. The federal toxicology program concluded with moderate confidence that fluoride above that threshold lowers children's IQ, but explicitly stated the data were insufficient to judge the 0.7 mg/L level. At fluoridation levels, the strongest direct studies—a 26,820-student US analysis, a 357-person Australian cohort, and a 2,001-woman Canadian pregnancy cohort—found no harm and, in the US case, slightly better cognition. The claims of low-level harm rest on urinary biomarkers that mix all fluoride sources, not just water, and on extrapolation from higher-dose studies. A 2025 meta-analysis of 74 studies found a significant inverse association overall, yet when fluoride was measured in drinking water the negative association was significant only above 2 mg/L; only urinary measures showed effects at lower levels. A 2023 dose-response analysis found the inverse association was strongest in high-bias studies and absent in the single low-bias study, and its urinary estimate near the equivalent of 0.7 mg/L water did not reach statistical significance. Despite this unsettled science, regulators are moving: a federal court found unreasonable risk under chemical-safety law, EPA appealed and opened a review, and FDA warned against fluoride supplements for the youngest children. About 209 million Americans drink fluoridated water, so any genuine cognitive risk would be a population-scale harm, while an unwarranted reversal would forfeit a major dental-health benefit.
The Full Investigation
8 sections · 13 min read
Confirmed facts and attributed reporting read normally; only contested, unverified, or speculative sentences are highlighted. Hover any sentence for its grade and sources.
A public health staple, suddenly on trial
For most Americans, the fluoride in their tap water is invisible infrastructure - added since the mid-twentieth century to fight tooth decay and largely unquestioned. In 2022, roughly seven in ten people served by community water systems, about 209 million out of 289 million, received fluoridated water. That everyday arrangement is now the subject of courtroom fights, federal reviews, and duelling scientific camps, all circling one deceptively narrow question: does the fluoride at the dose Americans actually drink quietly harm children's developing brains?
The number at the centre of the dispute is 0.7 milligrams per litre - the concentration recommended for US community water fluoridation. The regulatory ceiling sits far higher: since 1986 the EPA has enforced a maximum contaminant level of 4.0 mg/L, a limit set to prevent crippling skeletal fluorosis rather than to protect cognition. Much of the alarming research, however, comes from places where natural fluoride runs well above the US target, which is why the debate hinges less on whether high fluoride can harm and more on whether that harm extends down to 0.7 mg/L.
This report works through what the peer-reviewed evidence shows at each dose, what it shows specifically at fluoridation levels, where the methodological fights lie, and how regulators have responded. The evidence base is unusually rich but sharply divided, and the grade of each finding - solid, single-sourced, or contested - is itself part of the story.
The studied doses mostly sit far above what Americans drink
The first thing that separates the alarming headlines from the American tap is arithmetic. Most of the research reporting cognitive harm was conducted where fluoride occurs naturally at levels well beyond the US standard. The Fluoride Action Network, an advocacy group that opposes fluoridation, compiles what is probably the largest such list: by its own count, 78 of 87 human IQ studies - covering some 29,130 children and 689 adults - reported lowered IQ from elevated fluoride, with the studies concentrated in China (47), India (14), Iran, Mexico, and Canada. That compilation is the group's own characterization, not an independent audit, and most of those Chinese and Indian settings are endemic high-fluoride regions with co-exposures to arsenic and iodine deficiency that differ sharply from a US municipal system.
Even within the literature that reports water concentrations, the exposures skew high. Grandjean and colleagues, in a 2019 review, reported that 13 of 18 studies with water-fluoride data found cognitive deficits below the EPA's health-based goal, but the average elevated level in those studies was 2.3 mg/L - more than three times the US target - and the lowest was 0.8 mg/L. That review also reported a benchmark dose lower confidence limit of roughly 0.2 mg/L or below, which would place a threshold beneath fluoridated water. Both figures rest on that single review, with no independent corroboration, and the benchmark estimate in particular is a modelling result rather than an observed effect.
The authoritative federal read lands higher. The National Toxicology Program concluded with moderate confidence that fluoride above 1.5 mg/L is associated with lower IQ in children. The American Society for Nutrition's rapid review reached a similar boundary, reporting that neurological and thyroid effects generally appeared above 1.5 mg/L, with no evidence of adverse effects below 1.0 mg/L in children or adults. So when independent sources are compared, they diverge in an interpretable way: those relying on water concentration cluster around 1.5 mg/L and above, while the lower thresholds come from urinary biomarkers and high-quality subsets - a distinction the next section pulls apart. The Annals of Global Health editorial, drawing on the 2025 meta-analysis, argues that inverse associations appeared below 1.5 mg/L for high-quality studies using both urinary and drinking-water measures - a claim that stands on one source and, on the water-measured half, sits in tension with the meta-analysis's own reading.
2012
- Choi/Grandjean publish meta-analysis of 27 cross-sectional fluoride-IQ studies, finding standardized mean difference of -0.45
2019-12-19
- Grandjean et al. publish Environmental Health review analyzing fluoride neurodevelopmental evidence, including ELEMENT and MIREC cohort findings
2023-03-15
- Veneri et al. publish dose-response meta-analysis in Environmental Research, reporting -3.05 IQ points per 1 mg/L water fluoride above 1 mg/L
2024-08-21
- National Toxicology Program releases final report concluding with moderate confidence that fluoride above 1.5 mg/L is associated with lower IQ in children, but insufficient data for 0.7 mg/L level
2024-09
- US District Court issues ruling on water fluoridation and unreasonable risk under TSCA
2024-11
- US District Court rules that water fluoridation at 0.7 mg/L poses unreasonable risk of reducing IQ in children under TSCA, orders EPA to initiate rulemaking
2024-12-18
- Australian longitudinal study publishes in Journal of Dental Research finding no negative association between fluoride exposure and cognitive development in 357 participants aged 16-26
2025-01-06
- JAMA Pediatrics publishes meta-analysis of 74 epidemiological fluoride-IQ studies finding β = -1.63 (95% CI: -2.33 to -0.93; p<0.001)
2025-01-17
- EPA appeals federal district court TSCA fluoride ruling
2025-04
- EPA announces review of scientific evidence surrounding fluoride in drinking water
2025-10-31
- FDA publishes report warning against orally ingestible fluoride supplements for children under age 3 or at low/moderate caries risk
2025-11-11
- Canadian MIREC study publishes in Environmental Health (BMC) finding no association between prenatal fluoride exposure and autistic behaviors in 2,001 mother-child pairs
2025-11-19
- Warren et al. publish Science Advances study analyzing 26,820 US high school students, finding fluoride exposure at 0.7 mg/L associated with modestly better cognitive performance
2025-12
- Advances in Nutrition publishes rapid systematic review of 58 studies finding no adverse non-dental health effects from community water fluoridation at 0.7-1.0 mg/L
2026-02-03
- IADR releases position statement critiquing MIREC and ELEMENT study methodologies, alleging confounding by tester assignment and inadequate clustering adjustment
Open: Whether the lowest-exposure water-fluoride studies (down to 0.8 mg/L) adjusted adequately for the co-exposures typical of endemic regions is not resolvable from the graded claims.; The benchmark dose estimate near 0.2 mg/L rests on a single review and has no independent replication in the dossier.
Can high-dose effects be drawn down to low doses? The dose curve says not cleanly
If most studies sit above 1.5 mg/L, the whole question becomes whether their effects can be projected downward to 0.7 mg/L along a straight line. The meta-analyses that have tried say the picture is messier than a simple slope. The most recent and comprehensive, a 2025 JAMA Pediatrics pooling of 74 studies across 12 countries, found a statistically significant inverse association overall - a coefficient of -1.63 IQ points, with a confidence interval running from -2.33 to -0.93 and a p-value below 0.001. That is a real signal in the aggregate.
But the same meta-analysis contained the crucial caveat that undercuts a clean extrapolation. When fluoride was measured in drinking water, the negative association with children's IQ was significant only above 2 mg/L and not below 1.5 mg/L; only when fluoride was measured in urine did the inverse association persist below 1.5 mg/L. Urinary fluoride is not the same quantity as water fluoride - it reflects total intake from tea, processed food, and dental products as well as water - so the two measures answer different questions and cannot be read off the same scale.
The dose-response work sharpens this. Veneri and colleagues in 2023 reported a substantially linear decrease of about 3.05 IQ points per 1 mg/L for water fluoride between 1 and 2 mg/L, steepening above 2 mg/L. Their urinary estimate near the equivalent of 0.7 mg/L water was weaker - about 2.15 points, but with a confidence interval of -4.48 to 0.18 that crosses zero, meaning it did not reach statistical significance. The point estimate leans negative; the statistics do not confirm it. Older work ran hotter: the 2012 Choi and Grandjean meta-analysis of 27 studies found a standardized difference equivalent to roughly 6.75 IQ points, with no clear publication bias. These estimates are not directly poolable - they use different metrics, dose ranges, and study sets across more than a decade - which is itself why the field looks divergent rather than convergent.
One finding in the Veneri analysis matters more than the headline number for the extrapolation question: the inverse association was strongest in high-risk-of-bias studies, and no adverse effect emerged in the single study judged at low risk of bias. That pattern - stronger signal where the methods are weaker - is precisely what would make downward extrapolation unreliable.
Open: No stratified meta-regression in the dossier tests threshold versus linear model fit specifically in the below-1.5 mg/L range using water-measured exposure.; Whether the older 2012 effect size reflects study overlap with later meta-analyses cannot be determined from the graded claims.
At the dose Americans actually drink, the direct evidence points the other way
The cleanest test is not a projection at all but a study of populations drinking water at 0.7 mg/L. On that terrain the largest evidence runs against harm. A November 2025 Science Advances study analysed 26,820 students from more than 1,000 US high schools and found that childhood fluoride exposure at fluoridation levels was associated with modestly better cognitive performance in secondary school, according to reporting from the American Dental Association and DOCS Education. The advantage faded to non-significance by around age 60, though the point estimates stayed positive. The ADA is a long-standing promoter of fluoridation, so its framing carries an institutional stake; the underlying student and school figures, however, are reported identically by two independent outlets.
A 2024 Australian longitudinal study of 357 participants aged 16 to 26 pointed the same way, finding early-childhood fluoride exposure was not negatively associated with cognitive development, with full-scale IQ scores equivalent across exposure groups. And a 2025 Canadian MIREC cohort of 2,001 pregnant women across 10 cities found no association between prenatal fluoride and autistic behaviours in preschoolers. That cohort's exposures overlap the fluoridation range: median maternal urinary fluoride was 0.43 mg/L, averaging 0.72 mg/L in fluoridated regions against 0.42 in non-fluoridated ones - overlapping distributions that make it hard to isolate water's contribution.
Against these sit the prospective cohorts that anchor the harm case, and they are not trivial. The Mexican ELEMENT study reported that each 0.5 mg/L rise in maternal urinary fluoride predicted a 2.12-point drop in full-scale IQ, with a confidence interval of -3.49 to -0.75 that excludes zero. The Canadian MIREC IQ work reported a 4.49-point loss in boys per 1 mg/L of maternal urinary fluoride. These are prospective designs with individual biomarkers - a strength - but they measure urinary fluoride, the total-intake proxy that cannot be pinned to water alone.
So the direct evidence at fluoridation levels genuinely diverges. Three large null-or-positive studies in fluoridated populations line up on one side; two prenatal urinary-biomarker cohorts and a non-significant dose-response estimate line up on the other. The rapid review of 58 studies reported no adverse non-dental effects at 0.7 to 1.0 mg/L, but rests on a single source. And the program regulators treat as authoritative, the NTP, declined to call it either way, holding that the data at 0.7 mg/L were simply insufficient.
Open: No large prospective US cohort in the dossier measures both water and individual urinary fluoride in the same 0.7 mg/L population with blinded IQ testing.; The ADA's characterization of 'two large US cohort studies' finding no lifetime effect rests on a single advocacy source [C-033].
The inconvenient question: how much of the harm signal is confounding and study quality?
If the strongest low-dose signal comes from weaker studies, the natural next question is whether the harm literature is trustworthy at all. Here the critique comes largely from parties with a stake in fluoridation, and it must be read as attributed argument rather than settled fact. All three of the following allegations rest on a single interested-party source: the International Association for Dental, Oral, and Craniofacial Research, a professional research body, argues that the MIREC study used six different IQ testers each assigned to a different city, making it impossible to separate fluoride's effect from city and tester effects. The same statement contends that MIREC and ELEMENT did not adequately account for clustered observations - children nested within cities - producing misleading precision and potentially spurious associations. IADR also states mean IQ was 108 in both fluoridated and non-fluoridated MIREC communities, arguing IQ was not a public health concern in that sample. The IQ-108 figure does not appear in the 2025 MIREC autism paper, suggesting it references an earlier study not directly documented here.
The methodological worry is not purely partisan, though. The confirmed Veneri finding - inverse associations strongest where bias risk was highest, and absent in the one low-bias study - is exactly the pattern critics point to. That is independent of any advocacy framing, and it is the single strongest piece of the quality-critique case.
The American Fluoridation Society, which promotes fluoridation, adds two points. It states that the NTP's final report removed the original assertion that fluoride is a presumed neurodevelopmental hazard - a change corroborated by a second source noting the earlier language was withdrawn. It also states that 62% of the studies NTP reviewed were published in journals not indexed by the National Library of Medicine, a proxy for weaker publication pathways; that figure, too, comes from a single advocacy source with no independent check. Taken together, the critique is credible in outline - study quality clearly modifies the fluoride-IQ signal - but its most specific claims are single-sourced allegations from organizations that favour fluoridation, and none should carry more weight than that attribution allows.
Open: No independent statistical re-analysis of the MIREC or ELEMENT datasets addressing the clustering and tester-confounding allegations appears in the dossier.; The 62% non-indexed-journal figure has no independent verification.; Whether equivalent bias scrutiny was applied to the null-and-positive studies is not addressed by any source here.
Regulators are moving faster than the science has settled
While researchers argue over dose curves, courts and agencies have acted - and their actions have been read, by advocates, as verdicts on the science they are not. The most consequential is legal. According to the Congressional Research Service, a nonpartisan legislative body, a federal district court in November 2024 found that fluoridation at 0.7 mg/L presents an unreasonable risk under the Toxic Substances Control Act and ordered EPA to begin rulemaking; EPA appealed on 17 January 2025. The Annals of Global Health editorial placed the ruling in September 2024; CRS, the more authoritative T1 source for US legal events, placed it in November. A court finding of legal risk under a chemical-safety statute is not a finding of scientific consensus.
The EPA's own technical work has inched toward a lower ceiling. The CRS reports that the agency's 2024 six-year review used a reference dose of 0.08 mg/kg/day to calculate a potential maximum contaminant level of 0.9 mg/L - closer to fluoridation levels than the standing 4.0 mg/L limit, though still above 0.7. In April 2025 the EPA announced it would review the fluoride science, according to the Annals editorial, a single-sourced report with no contradicting account.
The FDA moved on a related front. In an October 2025 report, it warned against orally ingestible fluoride supplements for children under three or at low-to-moderate caries risk, noting the supplements deliver concentrations more than double community fluoridation levels. That targets pills, not tap water, and its logic - concern at doses well above 0.7 mg/L - is consistent with, rather than a repudiation of, community fluoridation. In short, the regulatory machinery is responding to genuine uncertainty and litigation, not to a new scientific finding that 0.7 mg/L causes harm.
Open: The EPA April 2025 review announcement and the 0.9 mg/L potential MCL calculation each rest on a single source.; The dossier contains no documented CDC or WHO position statement on the current cognitive-harm evidence.
Competing explanations, weighed against the evidence
Four explanations compete to account for the whole pattern, and the evidence does not distribute among them evenly.
The first holds that fluoride at 0.7 mg/L itself causes measurable cognitive harm. Its support is thin and indirect: a non-significant negative dose-response estimate at the urinary equivalent of 0.7 mg/L and a court's legal risk finding. Against it stand the NTP's insufficient-data conclusion, the 26,820-student US study, the 58-study review, the Australian cohort, and the MIREC autism cohort. On the graded evidence this explanation stands weak - no confirmed, direct study demonstrates harm at 0.7 mg/L.
The second holds that harm is real but confined to elevated exposures, generally above 1.5 mg/L, with meta-analytic signals reflecting high-dose studies. This is the best-supported reading: the NTP boundary, the nutrition review's 1.5 mg/L threshold, the water-measured meta-analysis result significant only above 2 mg/L, and the null-or-positive low-dose cohorts all converge on it. Its friction points are the below-1.5 mg/L urinary findings and the lowest-exposure water studies at 0.8 mg/L.
The third is subtler and may reconcile the other two: effects appear below 1.5 mg/L when measured by urinary biomarkers - which capture total body burden - but not by water concentration alone. The JAMA meta-analysis's water-versus-urine split, the ELEMENT and MIREC urinary cohorts, and the Veneri urinary estimate all fit this pattern, which is contradicted mainly by the no-effect-below-1.0 mg/L characterization. It remains plausible rather than confirmed, because urinary fluoride cannot be tied to water fluoridation specifically without measuring both in one population.
The fourth holds that reported associations stem from confounding, bias, and clustering errors rather than causal fluoride effects. The confirmed Veneri bias-gradient finding and the null high-quality studies support it; the significant JAMA aggregate, the older Grandjean pooling, the Veneri water-dose slope, and the ELEMENT cohort resist it. It is plausible but not decisive: study quality clearly modifies the signal without erasing every trace of it. The discriminating study every hypothesis points toward is the same - a large prospective US cohort at 0.7 mg/L measuring both water and individual urinary fluoride, with blinded IQ testing and full confounder adjustment - and it has not been done.
Open: No individual-participant-data meta-analysis restricted to water fluoride below 1.5 mg/L exists in the dossier to test heterogeneity and effect modification directly.
What the evidence forces, and what it leaves open
The evidence forces a narrow but firm conclusion: no confirmed, direct study demonstrates cognitive harm from fluoride at 0.7 mg/L, and the claims that it does rely on extrapolation from higher doses or on urinary biomarkers that cannot be tied to water fluoridation alone. The confirmed findings - the NTP's 1.5 mg/L boundary, its insufficient-data verdict at 0.7 mg/L, the water-measured meta-analysis significant only above 2 mg/L, and the null-or-positive US and Australian cohorts - all point the same way at the dose that matters.
That is not the same as a clean bill of health. The evidence base is genuinely divergent, and honesty requires holding two facts together: high fluoride can lower IQ, and the question of whether US fluoridation levels do remains, by the toxicology program's own account, unresolved. The prospective urinary-biomarker cohorts are real signals that a purely reassuring reading would have to explain away, and the non-significant negative trend near 0.7 mg/L is a point estimate leaning the wrong direction even if it does not clear the statistical bar.
Weighing the graded evidence, the most defensible position is the dose-dependent, low-dose-unresolved reading. Harm is confirmed above roughly 1.5 mg/L; at 0.7 mg/L the direct evidence leans toward no harm, but the data are too sparse and too entangled with measurement type to close the case. Regulatory motion - a court's risk finding, an EPA review, an FDA supplement warning - reflects precaution and litigation, not a scientific demonstration that community fluoridation harms cognition. Speculatively, and labelled as such, if a large US cohort measuring both water and urinary fluoride at 0.7 mg/L were run to completion, the pattern in the existing data suggests it would most likely find a small-or-null water-measured effect alongside a detectable urinary-measured one - which would vindicate the biomarker hypothesis without indicting the tap. That scenario is reasoning from the current evidence, not a finding.
Why it matters
About 209 million Americans drink fluoridated water, so any genuine cognitive risk at 0.7 mg/L would be a population-scale harm, while an unwarranted reversal would forfeit a major dental-health benefit. A federal court has already found unreasonable risk and ordered rulemaking, the EPA is reviewing its standards, and the FDA has restricted supplements - decisions that will shape policy for millions before the science is settled. Getting the dose question right, rather than conflating high-fluoride harm with fluoridation-level harm, is what separates evidence-based policy from precaution driven by extrapolation.