Where Is Glyphosate Found? Food, Water, and What Urine Testing Shows

This is the exposure-route article — part of the wider picture of everyday chemical exposure this library documents. Whether glyphosate at these levels does anything to a human body is a separate question with a genuinely unresolved answer; that argument, and where IARC and the regulators part company, is covered on its own page. What follows is the narrower and more answerable question: which foods, which water, which activities — and what has actually turned up when people measured it in human urine.

The famous route is the Roundup ReadyCrops genetically engineered to survive glyphosate application, introduced by Monsanto in 1996 — soybeans, maize, cotton, canola and sugar beet route: crops engineered to survive being sprayed, so the field can be cleaned of weeds without killing the harvest. That spray goes on mid-season, weeks to months before the combine arrives, and the crop it protects mostly reaches people indirectly — as refined oil, animal feed, syrup or ethanol. Glyphosate is a small, highly polar, water-soluble molecule. It does not partition into refined oil. The most visible route delivers less to the plate than its reputation suggests.

Pre-harvest desiccationSpraying a herbicide onto a mature crop shortly before harvest to kill it and speed even drying. Standard on oats, wheat, barley, lentils, chickpeas and dried peas in the US, UK and Canada. Prohibited as a desiccant use in the EU's 2023 glyphosate renewal. is the mirror image. The application happens three to seven days out. The target is the food itself, not the weeds around it. And the crops treated this way — small grains and pulses — are eaten close to whole, with no oil-refining step to leave the residue behind. Small share of the tonnage; large share of what ends up in a bowl.

You will see a specific number attached to this: that desiccation is around 2% of agricultural glyphosate use but more than half of dietary exposure. It is worth knowing where that came from. It traces to Charles Benbrook — the agricultural economist whose usage dataset underpins most of the volume figures in this field Benbrook 2016 — as a remark quoted in a 2016 magazine feature, not as a published calculation with a method attached. The direction is well supported by residue monitoring, which is what the rest of this article is about. The precise split is an estimate that has been repeated often enough to read like a measurement. Treat 2% / 50% as shorthand for the pattern, not as a measured quantity.

Start with the largest testing programme in the world. For the 2024 monitoring year the EUEuropean Union analysed glyphosate in 14,607 food samples analysed for glyphosate across 28 reporting countries in the EU's 2024 monitoring year food samples across 28 reporting countries. In 98.2% of them nothing was quantifiable at all. In 1.6% a residue sat above the reporting limit but below the legal maximum, 0.2% exceeded that maximum, and after measurement uncertainty was applied 16 samples — 0.1% — were formally non-compliant, mostly buckwheat and other pseudo-cereals EFSA 2026.

Now Canada, which tested the same chemical and produced a headline that reads like a contradiction. The Canadian Food Inspection Agency ran 7,955 Canadian retail food samples tested for glyphosate between April 2015 and March 2017 retail samples between April 2015 and March 2017 and found detectable glyphosate in 42.3% of them, with a 99.4% compliance rate against Canadian limits Kolakowski et al. 2020. Within the agency's 2015-16 tranche the pattern is the one desiccation predicts: grain products at 36.6%, and bean, pea, lentil, chickpea and soy products at 47.4% of Canadian bean, pea, lentil, chickpea and soy product samples carried detectable glyphosate — the highest category in the survey, against 29.7% across the survey as a whole CFIA 2017. Infant cereal came in at 31.7%.

Both sets of numbers are correct. They are not measuring the same thing. Canada set out to find glyphosate — a method built for it, a low reporting limit, aimed at the food groups most likely to carry it. The EU figure is a compliance denominator spread across every commodity in the basket, with reporting limits set for enforcement rather than trace detection. A detection rate answers "is it there." A compliance rate answers "is it legal." Neither one answers "how much am I eating." Reading either as the other is the most common mistake made with this data, and it is made in both directions — by campaigners quoting detection rates as if they were exceedances, and by industry quoting compliance rates as if they were absence.

The US started looking late. The FDAUS Food and Drug Administration added glyphosate to its residue monitoring programme in fiscal 2016 — the first systematic federal testing for the most heavily applied agricultural chemical in the American food supply — and found it in 63% of corn samples and 67% of soybean samples, all below EPAEnvironmental Protection Agency tolerances, and in none of the milk or egg samples FDA 2019. The much larger USDAUnited States Department of Agriculture Pesticide Data Program still does not carry glyphosate in its routine annual panel. Advocacy laboratories filled that gap, and their oat testing is where most public awareness of this issue came from — real laboratory work, but with sampling designs that were never independently reviewed and internal benchmarks that are not regulatory thresholds. The government programmes above are the ones with published methods and stable denominators.

The USGSUnited States Geological Survey compiled 3,732 water and sediment samples from 38 US states collected 2001-2010 — the largest US assessment of glyphosate occurrence water and sediment samples collected across 38 states between 2001 and 2010. Glyphosate and its breakdown product AMPAAminomethylphosphonic acid — the main environmental degradation product of glyphosate, which persists in soil roughly three times longer than the parent compound were detected frequently in soils and sediment, ditches and drains, rivers and streams — and in precipitation, which is to say it comes back down in the rain. They were found less often in lakes, ponds, wetlands, soil water and groundwater Battaglin et al. 2014. It is the same runoff pathway that puts atrazine into Midwest surface water, and it behaves in the same seasonal way.

Then the regulatory limits, which is where the jurisdictions diverge most sharply. The US EPA sets a maximum contaminant level for glyphosate in drinking water of 700 µg/L US EPA maximum contaminant level for glyphosate in drinking water, set in 1992, established in 1992 from animal toxicity work done in the 1980s. The EU's Drinking Water Directive caps any individual pesticide at 0.1 µg/L EU per-pesticide drinking water limit — a uniform policy floor set near the historic limit of detection, not a glyphosate-specific health threshold. That is a 7,000-fold gap, and it is mostly not a disagreement about glyphosate toxicology. The EU number is a single uniform figure applied to every pesticide regardless of its properties — a policy position that pesticides should not be in drinking water at all, pegged near the historic limit of detection. The US number is substance-specific and derived from a hazard assessment. Comparing the two as if they were rival answers to one question is the most common error in this area. Where campaigners have sampled European surface waters directly, exceedances of 0.1 µg/L do show up — but surface water is not tap water, and the treatment step between them is the whole point.

Non-agricultural use — gardens, drives, paths, amenity land, rights of way, industrial sites — used to account for 34-42% of total US glyphosate use. By 2014 it was under 10%, not because people stopped spraying but because agricultural volume grew around it Benbrook 2016. In national tonnage the garden is a rounding error. In personal dose it is not, because you are standing next to the nozzle.

The Farm Family Exposure Study measured exactly this. Forty-eight farmers, their spouses and 79 children aged 4 to 18 gave urine samples around an application day. 60% of farmers had detectable urinary glyphosate on the day they applied it — Farm Family Exposure Study of the farmers had detectable glyphosate on the day they sprayed, geometric mean 3 ppb, maximum 233 ppb. Spouses: 4%, maximum 3 ppb. Children: 12%, and mostly the ones who had been present during spraying. The strongest single predictor among the farmers was not acreage or equipment — it was gloves. A 10 ppb geometric mean urinary glyphosate in farmers not wearing rubber gloves, against 2.0 ppb in those who did geometric mean without rubber gloves against 2.0 ppb with them Acquavella et al. 2004.

A 2024 biomonitoring study from the US National Cancer Institute reached the same shape of answer with a modern method. Farmers who had used glyphosate in the previous seven days carried a geometric mean roughly 5.5× that of non-farming controls, and those non-farming controls sat at ordinary general-population levels Chang et al. 2024. If you never spray it, spraying is not your route. If you do spray it, it is comfortably your largest one — and what you wear on your hands matters more than anything printed on the label.

The reference number comes from the US NHANESNational Health and Nutrition Examination Survey — the CDC's rolling national health survey, which banks urine samples for chemical biomonitoring programme. In the 2013-2014 cycle, glyphosate was detectable in 81% of Americans aged six and older had detectable urinary glyphosate in NHANES 2013-2014 of Americans aged six and older, at a population geometric mean of 0.411 µg/L. Children aged 6-11 ran higher than adults — 0.515 against 0.392 Ospina et al. 2022. Children eat more food per kilogram of body weight, and more of that food is cereal.

Extending the same measurements through 2018 changed the story in a way that rarely gets reported. Detection stayed high — between 70.0% and 81.7% depending on cycle, including in children as young as three — but the geometric mean fell 38% decline in US population geometric mean urinary glyphosate between the 2013-2014 and 2017-2018 NHANES cycles between 2013 and 2018, with the smallest declines in the youngest age groups Ospina et al. 2024. Germany's Environmental Specimen Bank, which holds banked 24-hour urine samples going back decades, shows the same arc from the other side of the Atlantic: of 399 samples analysed, 31.8% carried glyphosate at or above the reporting limit, but the fraction rose from 10.0% in 2001 to a peak of 57.5% in 2012 before falling back to 32.5% in 2014 Conrad et al. 2017.

What these concentrations mean is the point where honesty costs something. None of the systemic doses estimated in the Farm Family study — the most heavily exposed group in this entire picture, measured on the day they sprayed — approached the EPA reference dose Acquavella et al. 2004. That is the strongest card the industry position holds and it deserves to be stated plainly rather than buried. What it does not settle is whether the reference doses sit in the right place, which is precisely where the jurisdictions disagree again: the EU acceptable daily intake is 0.5 mg/kg EU acceptable daily intake for glyphosate, per kilogram of body weight per day of body weight per day, and the US EPA chronic reference dose is 1.75 — three and a half times higher, derived from overlapping animal studies read differently. A measurement well below a limit is reassuring in proportion to your confidence in the limit.

The cleanest intervention evidence is a diet swap. Fagan and colleagues moved US families onto an entirely organic diet for six days and measured urinary glyphosate and AMPA before and during. Mean glyphosate fell 70.9% drop in mean urinary glyphosate after six days on an entirely organic diet, AMPA fell 76.7%, and the drop was fast — near floor within three days, with adults and children responding almost identically Fagan et al. 2020. The study was funded by organic-sector organisations, which you should know before quoting it. The effect size is large, the mechanism is obvious given that food is the dominant route, and funding is not a plausible explanation for the whole of it.

Then a processing finding with an uncomfortable edge. Tittlemier and colleagues tracked glyphosate through wheat milling and bread production. 81% of the glyphosate in a wheat kernel sits in the bran, shorts and feed milling fractions — the parts removed to make white flour of the glyphosate mass in the wheat sat in the bran, shorts and feed fractions, with half of it in the outer 17% of the kernel. Bread made from straight-grade white flour carried roughly 4× less glyphosate than bread from wholegrain flour. Baking changed nothing at all — dough, fermentation and oven left the residue exactly where it was Tittlemier et al. 2021. So the effective processing step is refining out the fibre, which is the one thing you should not do for other reasons. Buy wholegrain, and buy it organic, rather than trading fibre for a residue that already sits below regulatory limits.

The shape of this answer is not the one most people arrive expecting. The engineered crops that made glyphosate famous are the smaller dietary route. The larger one is a harvest-timing convenience applied to plain oats, plain wheat and plain lentils — a practice the EU stopped in 2023 and the US, UK and Canada have not. Water is a minor route once it has been through a treatment works. The garden is a rounding error nationally and the dominant route personally, if you are the one holding the sprayer.

And the measurement trend runs against the headlines: US urinary levels fell 38% between 2013 and 2018, and the German time series peaked in 2012 and came back down. Exposure is widespread, it is genuinely measurable, and it is not increasing. Whether those levels matter is the contested question this article has deliberately left where it found it — the toxicology is unresolved and the regulators disagree in both directions. The precautionary version is cheap either way: change the four foods carrying most of it, and put gloves on if you spray.