The number on your water report, not the word fluoride, decides this. At the level added to United States drinking water — 0.7 mg/L — the frightening harms are unresolved rather than established. The child-IQ signal that made headlines is real, but the evidence for it sits above about 1.5 mg/L. Whether it reaches down to fluoridation levels is genuinely unsettled. Bone fractures rise only well above that line. The one unwanted effect actually proven at 0.7 mg/L is mild dental fluorosis — faint white flecks on tooth enamel, a cosmetic outcome. On the benefit side, fluoride toothpaste clearly prevents cavities, and adding fluoride to the water on top of that now buys only a small, uncertain extra.
So the three everyday choices come apart: keep using the fluoride toothpaste, drink the fluoridated tap water without alarm, and treat the water filter as a separate question that turns on what your water actually contains and at what concentration.
Two honesty notes frame everything below. First, the evidence at fluoridation-relevant doses is thin and low-certainty on both sides — the benefit is small and uncertain, and the serious harms are unproven rather than ruled out. Second, this is a contested literature. Two respected research groups pool almost the same studies yet reach opposite conclusions about whether the IQ harm survives once you keep only the best studies. A finding that flips depending on which quality tool you apply cannot anchor a threshold (Does the Fluoride IQ Harm Survive a Low Risk-of-Bias Restriction, Fluoride and Health).
Fluoride cuts cavities — but the water’s extra benefit shrank once toothpaste arrived
Fluoride prevents tooth decay; that much is not in dispute. The open question is how much community water fluoridation still adds once a population already brushes with fluoride toothpaste. The 2024 Cochrane review answers it by splitting the evidence at roughly 1975, when fluoride toothpaste became widespread. All 157 included studies were non-randomised (a fluoridation trial is logistically unfeasible), so every estimate is downgraded for that alone (Iheozor-Ejiofor et al., 2024).
In the modern era the added benefit of fluoridated water is about a quarter of a tooth, and the confidence interval includes no benefit at all. Starting fluoridation after 1975 gave a difference in decayed/missing/filled baby teeth of «approximately one-quarter of a tooth in favour of CWF; this effect estimate includes the possibility of benefit and no benefit» (MD 0.24, 95% CI -0.03 to 0.52; 2 studies, 2908 children; low certainty) (Iheozor-Ejiofor et al., 2024). The permanent-teeth estimate is weaker still (DMFT MD 0.27, 95% CI -0.11 to 0.66; very low certainty). No trial reported caries outcomes in adults (Iheozor-Ejiofor et al., 2024).
The large, familiar fluoridation benefit is a pre-toothpaste number, and quoting it today overstates the case. Before 1975, starting fluoridation cut baby-tooth decay by MD 2.10 (95% CI 1.71 to 2.49) — roughly nine times the contemporary estimate, and unlike it, statistically clear (Iheozor-Ejiofor et al., 2024).
The benefit shrank not because fluoride stopped working but because the comparison changed: the un-fluoridated population now gets fluoride from toothpaste, varnishes and sealants, so the marginal gain from also putting it in the water is diluted (Iheozor-Ejiofor et al., 2024). Judge fluoridation against the realistic alternative — a population already using fluoride toothpaste — not against no fluoride at all. The old, large benefit still transports to settings where decay is high and toothpaste is scarce, so this is a baseline-risk-dependent benefit, biggest where caries burden is worst (Iheozor-Ejiofor et al., 2024).
The one harm proven at the fluoridation dose is mild dental fluorosis, a cosmetic effect. At 0.7 mg/L, dental fluorosis of aesthetic concern was estimated at 12% (95% CI 8% to 17%), and fluorosis at any level at roughly 40% (95% CI 35% to 44%), at low certainty (Iheozor-Ejiofor et al., 2024). This is faint mottling of the enamel — a real unwanted effect, but not on the same tier as a lost IQ point or a hip fracture. It is the one harm that sits squarely at 0.7 mg/L rather than above 1.5.
The child-IQ concern is real above 1.5 mg/L and unresolved at fluoridation levels — the open question
This is the arm the fabric holds open. The United States National Toxicology Program’s 2024 systematic review concluded, at moderate confidence, that fluoride exposures above the WHO drinking-water guideline of 1.5 mg/L are «consistently associated with lower IQ in children» (National Toxicology Program, 2024). Below that level the review does not find a settled effect — associations at lower exposures remain unclear — and it explicitly declined to address whether fluoride at the 0.7 mg/L US dose affects IQ (National Toxicology Program, 2024). So the honest below-1.5 verdict is insufficient evidence — not a demonstrated null, and not a demonstrated harm. The US fluoridation dose, 0.7 mg/L, sits inside that unresolved zone. A bare fluoride lowers IQ with no exposure bound overstates the source.
The 1.5 mg/L line marks where the confident evidence sits, not a proven safe threshold. NTP chose the WHO 1.5 mg/L water guideline as its reference because no fluoride-specific safety guideline for total exposure exists — most of the high-quality studies sampled populations above it (National Toxicology Program, 2024). To read 1.5 as the point where a dose-response curve bends mistakes the edge of the data for a feature of the curve (The Underivable Optimum). harm below 1.5 is unclear is a statement about how thin the low-dose evidence is, not about where fluoride becomes safe.
The peer-reviewed dose-response splits by how exposure was measured. NTP’s companion meta-analysis, now published as Taylor 2025, found an individual-level decrease of 1.63 IQ points per 1 mg/L of urinary fluoride (95% CI -2.33 to -0.93). The estimate tightened to 1.14 points (95% CI -1.68 to -0.61) when restricted to low-risk-of-bias studies (Taylor et al., 2025).
But the two exposure metrics diverge at low dose. Measured by water concentration, the dose-response goes null below about 2 mg/L — the estimate below 1.5 mg/L is even faintly positive. Measured by urinary fluoride, which captures total intake from all sources, the inverse signal persists down below 1.5 (Taylor et al., 2025). Taylor attributes the water-based null to thin, low-contrast data at low dose, not to a safe threshold, so the low-dose reading is exposure-metric-dependent, and the persisting urinary signal rests on the noisier spot-urine measure (Taylor et al., 2025).
A per-mg/L shift sounds small, but the authors note that a 5-point population-wide drop in IQ would nearly double the number of people classified as intellectually disabled — small individual shifts matter at the distribution tails (Taylor et al., 2025).
The two best meta-analyses disagree on whether the harm survives a quality restriction
The load-bearing uncertainty is not the data — it is which risk-of-bias tool you trust. NTP and Taylor (one research stream, not two independent lines) report that the inverse association holds across risk-of-bias determinations and tightens in their 19 low-risk-of-bias studies (National Toxicology Program, 2024). An independent Italian group (Veneri 2023), pooling almost the same primary studies with a stricter confounding tool, found the opposite: the association vanishes in the highest-quality studies (low-risk-of-bias MD +1.11, 95% CI -0.67 to 2.89, non-significant), with a «general trend towards weaker or null associations in the most carefully conducted studies» (Veneri et al., 2023).
The clash traces to a single flagship cohort (Bashash/ELEMENT): one tool rates it low-risk, the other high-risk. So the reassurance that it holds in the best studies is instrument-dependent, not a property of the evidence (Does the Fluoride IQ Harm Survive a Low Risk-of-Bias Restriction).
Both sides are contested (Veneri’s null rests on a single study), so this lowers confidence without resolving it. This is why the sub-1.5 mg/L verdict stays open.
Where the evidence was collected, and whether it transports to your tap
Most of the high-quality evidence comes from high-fluoride regions far above US fluoridation. The studies sampled near fluoridation levels are North American pregnancy cohorts, and they carry the transportability question — do they reach a 0.7 mg/L tap-water setting? Their signals exist but are fragile and conditional: Bashash (Mexico) found -2.50 full-scale IQ points per 0.5 mg/L maternal urinary fluoride (95% CI -4.12 to -0.59); Green (Canada) found an effect in boys only (-4.49 per 1 mg/L, 95% CI -8.38 to -0.60; girls and the combined estimate null); Till (Canada) found it in formula-fed but not breastfed infants (National Toxicology Program, 2024). These are prenatal and infant exposures via the mother — not the same claim as 0.7 mg/L in tap water lowers a child’s IQ.
Cutting the other way, the Yu (China) cohort found an effect only between 3.4-3.9 mg/L with nothing significant below (National Toxicology Program, 2024), and the New Zealand Dunedin cohort — followed to age 38 under community fluoridation — found «no clear differences in IQ» (National Toxicology Program, 2024). Fluoride does escape the worst of Measurement Error in Dietary Assessment, because urinary fluoride is a biomarker rather than a food-recall estimate. But area-level water assignment and unmeasured non-water sources still push a true low-dose gradient toward the null.
Bone fractures rise only well above the fluoridation dose
At the fluoridation dose there is no fracture signal; the risk turns up above roughly 1.5 mg/L. The first dose-response meta-analysis of this arm (Mazzoli 2025) found fracture risk flat at low exposure and rising above about 1.4 mg/L: RR 1.00 at 1.2 mg/L (95% CI 0.80-1.24), 1.06 at 2.0 mg/L, 1.19 at 3.0 mg/L, and 1.35 at 4.0 mg/L — but every one of these overall estimates has a confidence interval that crosses 1.0 (Mazzoli et al., 2025). This is a positive trend, not a per-dose demonstrated harm, and it is low certainty. The studied range runs to about 4 mg/L; no pooled estimate exists above that, so the curve’s apparent top is the sampling edge, not a real feature.
One stratum reaches significance earlier: post-menopausal women. In the pre-specified females-over-50 analysis the risk rose sooner and reached significance at a fluoridation-adjacent dose: RR 1.26 (95% CI 1.10-1.46) at 1.0 mg/L. That is the only pooled estimate the paper reports whose interval excludes 1.0 (Mazzoli et al., 2025). The candidate mechanism is that oestrogen loss accelerates bone turnover and may release stored fluoride, plausibly leaving the post-menopausal skeleton more susceptible (Menopause and the Shifting Levers). Treat this as a stratum hypothesis on low-certainty, single-lab data, not an established effect.
Do not read a rise in bone density as a benefit — density and fracture come apart here. Fluoride can raise bone mineral density at some sites while fracture risk also rises, because «BMD only captures bone quantity (density), not bone quality» — the architecture and material properties that actually govern fractures (Mazzoli et al., 2025). So density is a bad surrogate for the outcome that matters; steer on the fracture endpoint (Surrogate Outcomes).
The curves also show an apparent protective dip at low fluoride (a U-shape), but that lower arm is not believed. The source itself flags it as a likely artifact of sparse low-dose data, and the signal actually weakens when the worst studies are removed (fragility-fracture RR 0.93, 95% CI 0.72-1.21 across 0.8-1.5 mg/L) — the direction that argues against a real effect (The U-Shaped Association Artifact) (Mazzoli et al., 2025).
What’s actually in treated tap water — fluoride is not the only additive, and the others are a different question
When people ask should I filter my tap water?, fluoride is rarely the biggest thing in it — and this is not a fluoride effect. Chlorinating water to kill pathogens generates disinfection by-products, of which trihalomethanes are the regulated marker. A gold pooled analysis (Costet 2011) found a modest bladder-cancer signal in men only: odds ratio 1.47 (95% CI 1.05-2.05) at the highest exposure band versus the lowest, with no association in women and — importantly — no association via drinking (ingestion). The male signal instead tracked showering, bathing and pool use — skin and inhalation routes (Costet et al., 2011). A drinking-water filter therefore targets the route that was null.
Across the EU, these by-products were estimated to account for about 4.9% of bladder cancers (Evlampidou 2020), but this is a population-level, structural lever — water-treatment optimisation — and it is dwarfed in the same data by smoking (odds ratio 3.91 for current smokers) (Evlampidou et al., 2020) (Costet et al., 2011). Causation is not established, and this belongs on the what’s in treated water ledger, kept firmly separate from the fluoride question (Disinfection Byproducts and Bladder Cancer).
Where these doses sit, and who is affected differently
The whole appraisal turns on four reference points on the fluoride axis, and the same 1.5 mg/L line keeps recurring. Read the number off your water report against these:
- 0.7 mg/L — the dose added to US community water. Here the caries benefit is small and low-certainty, mild dental fluorosis is the one established effect, and the serious harm arms are insufficient-evidence.
- 1.5 mg/L — the WHO water guideline and NTP’s higher-exposure reference. Both the child-IQ harm and the overall fracture trend begin at roughly this line — while the marginal caries benefit has already flattened. This coincidence is the fabric’s core finding: the same exposure that turns on two harm arms is where the extra benefit runs out.
- 2.0 mg/L — the US EPA’s non-enforceable secondary limit (severe dental-fluorosis protection).
- 4.0 mg/L — the EPA’s enforceable limit, set for skeletal-fluorosis protection; the fracture trend is strongest here.
(National Toxicology Program, 2024) (Mazzoli et al., 2025)
Your total fluoride dose is not the water number. Even on optimally fluoridated water, intake varies with toothpaste use, dental products, and drinks such as black tea, so the water concentration under-counts the real dose (National Toxicology Program, 2024). A small US stratum matters most here: more than 2.9 million US residents drink community or private water above 1.5 mg/L (Taylor et al., 2025), and roughly 0.59% are on community systems naturally at or above that line — the group for whom the harm arms are live (National Toxicology Program, 2024). Life stage sharpens it further. The IQ signals near fluoridation levels are prenatal and infant exposures, so a pregnant woman on a high-fluoride private well faces a different calculus than a healthy adult; and post-menopausal women appear more fracture-susceptible earlier.
What to do: drink the water, use the toothpaste, and check the filter question separately
The three everyday decisions have different answers, and they hinge on your water number, not on whether fluoride is good or bad.
- Use the fluoride toothpaste. Topical fluoride is the clear, uncontested caries benefit, delivered at the tooth surface rather than swallowed (Fluoride and Caries Prevention). Nothing in the harm evidence bears on brushing with it.
- Drink the fluoridated municipal water without alarm — but know your number. At 0.7 mg/L the caries benefit is small in the toothpaste era, the only proven downside is faint cosmetic fluorosis, and the serious harms are unresolved rather than established. If your water is a private well or a naturally high-fluoride system, test it. Above about 1.5 mg/L the IQ and fracture arms become live and mitigation (defluoridation or an alternative source) is genuinely on the table — most so during pregnancy and for post-menopausal women (Fluoride and Child IQ, Fluoride and Bone Health).
- Treat the filter as a separate question. A drinking-water filter is a reasonable choice at genuinely high fluoride (>1.5 mg/L). For disinfection by-products it does little, because the bladder-cancer signal ran through skin and inhalation, not drinking (Disinfection Byproducts and Bladder Cancer). At normal municipal fluoride it solves a problem the evidence has not shown you have.
One axis this appraisal does not weigh: whether adding a substance to everyone’s water without individual consent is acceptable is a question about autonomy, not health. It is a real consideration and yours to weigh; the evidence here cannot settle it, and this page does not price it against the health findings.
The dose decides — so read your water number before you decide anything
Stop asking whether fluoride is safe and start asking at what concentration. For the ordinary reader on fluoridated municipal water at 0.7 mg/L, the practical verdict is calm: keep the toothpaste, keep drinking the water, and skip the filter unless your water is genuinely high in fluoride. The famous IQ harm is real above about 1.5 mg/L and honestly unresolved at fluoridation levels. The signal even flips direction depending on which quality tool a meta-analyst applies — exactly why no one should report it as settled in either direction. Fractures rise only well above the fluoridation dose. The whole menu at 0.7 mg/L is small effects held at low certainty, on both sides.
The one group who should act now is anyone drinking water above ~1.5 mg/L — test it, and consider an alternative source, especially in pregnancy. The open question the field has not closed is whether the harm reaches down to fluoridation levels; a US cohort at fluoridation-relevant exposure would help settle it, and none yet exists (Fluoride and Health).
Mandatory caveats
- Appraise, do not prescribe. This is a general, population-level appraisal, not medical advice; testing your water, selecting a filter or alternative source, and managing exposure in pregnancy or for a specific medical condition are decisions that require information this document does not hold.
- Open loop. This wiki grades internal coherence and fidelity to its sources — never whether a recommendation actually improves outcomes in the world. No step here tests a claim against a realized outcome, so a clean appraisal is not a validated result.
- Health axis only. This cut weighs fluoride on caries, child cognitive development, and bone/fracture outcomes; the autonomy of mass water supplementation, and cost and equity, are real considerations and yours to weigh, not appraised here.
- A general appraisal, applied per person. Your water’s fluoride concentration, your life stage (pregnancy, infancy, after menopause), and your baseline caries risk decide the individual weighting — at the end, not the top.
Evidence box
Question ’What does the evidence show about fluoride at everyday exposure levels — fluoridated drinking water and fluoride toothpaste — on each patient-important outcome (dental caries, bone and fracture, child cognitive development), and how does each effect vary with dose, life stage, and baseline caries risk — direction, magnitude, for whom, how certain?‘ Evidence included 7 sources — 6 gold, 1 high Overall certainty Low (see Rating Certainty of Evidence) Source-selection note All sources are gold or high tier. Last updated 2026-09-09 · Independently reviewed: No · Full edit history