Spinach Has Five Times the Cadmium of Potatoes. Your Potato Serving Has More.

We compared laboratory food tests with portions people actually eat. Some rankings reverse when you switch from a kilogram to a serving. Others become misleading unless you know exactly which chemical was measured. You can explore the underlying Food Market Atlas yourself.

A kilogram of raw spinach contained about five times as much cadmium as a kilogram of skin-on baked potato in US government food tests. That seems like a clear result.

Then we looked at the portions people reported eating. The spinach portion was about a handful of leaves (25 g); the potato was roughly one large baked potato (285 g). When we multiplied each food’s laboratory average by its observed eating-day portion, the comparison flipped: about 5.7 micrograms of cadmium in the spinach, against 12.9 in the potato.

Neither food had changed. Only the question had. The difference is not a health ranking: we combined measurements from two separate datasets, and neither tells us how often an individual eats either food.

The same trap appears in less obvious guises. Baked salmon seems to have almost six times as much total arsenic per portion as cooked white rice. But total arsenic includes chemicals with very different biological effects. And sometimes what looks like a precise concentration is mainly a decision about how to handle results the laboratory could not detect.

Explore the Food Market Atlas · compare 165 foods by portion ↓

In brief

• Concentration is not a portion. Raw spinach had more cadmium per kilogram in our FDA data, while the observed baked-potato portion contained more in a modelled comparison.
• The chemical form matters. Total arsenic in fish and inorganic arsenic in rice cannot be read as interchangeable toxic doses.
• Not detected does not mean zero. Explore the source records, sample counts and uncertainty rather than judging a food by one coloured bar.

We built the Food Market Atlas by connecting FDA and European food-monitoring results with observed US portions. The underlying dataset contains 943 food entries and 9,248 derived food–analyte profiles. The interactive serving comparison offers the 165 foods with matched observed eating-day portions; the other 778 entries remain available in the downloadable data. We conducted no new chemical tests. Our contribution is to bring existing records together, make their denominators visible and show which comparisons cannot bear the weight of a safety claim.

How we compared laboratory results with real portions

To put the laboratory results on a plate, we combined FDA food measurements from 2018–2022 with eating-day portions reported by US adults in NHANES. The raw spinach serving was 25 g; the skin-on baked potato serving was 285 g. These medians came from different groups of people, not a person’s daily routine.

Preparation matters: these figures describe raw spinach and baked potato with skin. Cooking and water loss may alter both concentration and portion weight. For mass concentrations, 1 µg/kg equals 1 part per billion (ppb). The table extends the comparison to familiar foods.

Source: FDA Total Diet Study 2018–2022 (44 laboratory composites per food) × NHANES dietary recalls 2021–2023 (weighted median portions among consumers; spinach n = 106, potato n = 34). Derived comparisons, not measured personal doses or a ranking of health risk. Values rounded to two significant figures.

Search among the 165 foods with observed eating-day portions to compare their modelled amounts. The downloadable derived dataset also includes 778 foods without matched portion estimates. These measurements are not a food-safety ranking; results from different markets are not averaged, and “not detected” does not mean zero.

The problem with the word “arsenic”

The second trap is subtler. With salmon and rice, the denominator is not the problem. The chemical name is.

Our calculations put total arsenic at roughly 47.5 micrograms in a 140-gram serving of baked salmon (about 5 oz), compared with 8.0 micrograms in a 158-gram serving of cooked white rice and 10.3 micrograms in a 147-gram serving of brown rice.

If those were all measurements of the same hazardous chemical form, the comparison could be informative. They are not.

Arsenic in seafood often includes arsenobetaine, but marine fish can also contain arsenolipids and other organoarsenic compounds, whose toxicology differs and is not always fully characterised. Rice can contain inorganic arsenic, alongside other forms. Total arsenic is not a toxic-equivalence measure. Our atlas has no matching inorganic-arsenic measurement for baked salmon, so the two foods cannot be ranked for that specific exposure.

But the FDA atlas does have selected inorganic-arsenic measurements for rice. Six analysed white-rice composites average about 52 µg/kg inorganic arsenic, or 8.3 µg for a modelled 158 g portion. Separately, 44 composites average about 50 µg/kg total arsenic, or 8.0 µg per portion. These were different, selectively analysed groups, so the slightly higher inorganic mean does not violate chemistry: they are not paired results and must not be subtracted. FDA specifically conducts arsenic speciation in selected rice-containing foods. Six composites are informative, not a representative national exposure estimate.

One useful action does not require a personalised risk score: the FDA advises that cooking rice in plenty of water, about 6–10 parts water to one part rice, then draining the excess, can reduce inorganic arsenic by roughly 40–60% depending on the rice. The trade-off is real: this also removes roughly 50–70% of some added iron and B vitamins from enriched rice. Simply rinsing rice before cooking has much less effect on arsenic. A varied selection of grains remains sensible, especially for children.

The same element can travel through food in different chemical forms, with very different biological consequences. That is why EFSA assesses inorganic arsenic separately from organic compounds.

Nor should the salmon example be confused with mercury. In the same FDA dataset, baked salmon averaged about 20 micrograms of total mercury per kilogram, equivalent to 2.8 micrograms in a 140-gram portion. Mercury exposure from fish depends strongly on species and, particularly, methylmercury content. The FDA and EPA offer species-specific fish advice, including special recommendations for pregnancy and childhood. Salmon is among their lower-mercury choices.

The useful question is not “Which food contains the most chemicals?” It is “Which substance, in which form, in how much of the food I actually eat?”

Sometimes “zero” means “we couldn't measure it”

The other difficulty lies at the low end of the scale.

Across 5,225 FDA results for each of the substances in our 2018–2022 analysis, approximately 86% of lead measurements and 92% of mercury measurements were reported below their respective laboratory thresholds. That is very different from demonstrating that those foods contained none of either element.

A result below the threshold may reflect a tiny concentration, a genuinely absent substance, or simply a measurement method unable to distinguish the two. Assigning every such result a zero can make one product appear remarkably clean. Assigning every result the laboratory limit can overstate its estimated average.

Consider chromium in a baked potato with its skin. Across 44 FDA composite results, alternative treatments of non-detects yield a mean of 2.8–51 µg/kg, an eighteen-fold span. Only 2 of 44 composite results had detectable chromium. At the 285 g portion, the corresponding modelled estimates are 0.80–14 µg. The difference reflects the handling of results below the reporting threshold, not the observed range of individual potatoes.

These two numbers are lower- and upper-bound substitutions, not a confidence interval or the minimum and maximum concentrations in potatoes on sale. Treating non-detects as zero in the lower calculation is not evidence of zero chromium in the food. More sensitive measurements are needed to narrow the uncertainty.

This is why the Explorer must never colour every food green, amber or red on the basis of a single contamination score. A useful dataset should admit when it cannot rank two foods at all.

How much do results vary between samples?

An average also hides dispersion. Among the 44 FDA composite results, the middle 80% of cadmium concentrations extended from approximately 100 to 368 micrograms per kilogram for raw spinach, and 25 to 64 for baked potato with skin.

But the word composite matters. For regional foods, the FDA may combine purchases from three cities into one prepared laboratory sample. Germany's MEAL Study also analyses pooled foods. Such methods are valuable for estimating what a diet contains on average. They can also smooth away the extremes of individual items.

So the ranges above describe the tested composites. They are not odds for the next bag of spinach or potato you buy. And a confidence interval around a laboratory average, even when available, would answer a different question from the spread between products on sale.

A composite is a laboratory unit, not a particular item on a shelf. That is what those percentiles describe.

What Europe adds, and what it doesn't

The German BfR MEAL Study measured contaminants in 356 commonly eaten foods, generally prepared as consumed. The additional European data come from national monitoring records submitted to EFSA by the Danish National Food Institute (DTU) and the Netherlands National Institute for Public Health and the Environment (RIVM). Those data include targeted controls and supply-chain sampling, not necessarily representative retail baskets. A product measured in the Netherlands may have been grown or manufactured elsewhere.

It does, however, complicate the averages. Germany's prepared-food composites, US regional composites and European market-control samples were collected for different purposes, in different years and with different food definitions. Some products have many results; others only a handful. Some measurements concern retail food, while others come from earlier points in the supply chain.

We therefore show each market separately. Where food identity or preparation cannot be matched confidently, the Explorer should say so. We do not calculate a league table of countries with the “cleanest food” or imply that equal numbers of laboratory records represent equal shares of supermarket sales.

The European evidence opens new comparisons. It does not tell us which country sells the safest food.

What should you do with any of this?

Start by resisting the temptation to remove a nutritious food because of one large-looking number. A trace concentration does not establish a health risk, while a high-quality diet has benefits a contaminants table cannot capture.

One important exception concerns people who are sensitised to nickel. EFSA’s 2020 nickel assessment uses a chronic tolerable daily intake of 13 µg/kg body weight, but also identifies acute flare-ups of systemic contact dermatitis after oral nickel exposure. The recurring-meal percentage bar in the Explorer does not assess that acute reaction.

What does help is to make decisions at the right level of specificity. Follow fish guidance by species, rather than avoiding seafood. If rice is a staple, consider a variety of grains, especially when feeding young children. And when you read a contamination headline, check whether it reports a concentration per kilogram, the amount in a realistic serving or an estimate of repeated exposure over time.

Cadmium illustrates why that last distinction matters. EFSA established a tolerable weekly intake of 2.5 micrograms per kilogram of body weight. For a 70-kilogram adult, that is 175 micrograms across an entire week from the whole diet. One modelled 285 g baked-potato portion contains approximately 12.9 µg. If that portion were eaten every day, the potato alone would contribute about 90 µg per week, or 52% of the weekly benchmark, before counting any other food. The new Explorer compares repeating two-food combinations. For example, its FDA/NHANES model estimates 27.7% of the weekly cadmium reference for rice plus raw spinach versus 15.5% for rice plus boiled broccoli, provided those specific eating-day portions were repeated every day. The rice portion stays fixed while the two vegetable portions differ, because each comes from a separate eating-day median. These are illustrative substitutions, not nutrition-equivalent meals or measured habitual diets. Other foods consumed each day could add considerable cadmium exposure, so the coloured bars show only the two chosen ingredients, never overall safety.

For lead, a percentage of a safe weekly limit would be misleading because no safe level has been identified. The FDA nevertheless uses interim reference levels of 2.2 µg/day for children and 8.8 µg/day for women of childbearing age. They are exposure-management benchmarks, not assurances of harmlessness or a general adult safe-intake limit. For arsenic, chemical form matters. For mercury, fish species and life stage matter. A universal safety score would hide these distinctions.

A number is not a verdict

The surprise is not that our food contains unwanted substances. Regulators have measured them for decades. It is how easily a seemingly definitive figure changes meaning when we move from a kilogram to a portion, from total arsenic to its chemical forms, or from “not detected” to an unknown concentration below a threshold.

We can calculate the estimated amount of a measured contaminant in many familiar portions. We can show how much the sampled foods differed. We can also show where the evidence stops.

Our atlas will not tell you to love spinach or fear potatoes. Its more useful gift is a better question: what, exactly, was measured, and what can that measurement tell us?

How we did this

Mike's Balance harmonised public laboratory and consumption data for Food Market Atlas v1.0. The US core uses the FDA Total Diet Study (2018–2022 fiscal years); observed US portions come from NHANES. European measurements include the BfR MEAL Study, Danish national monitoring and Dutch national monitoring; European food consumption was checked against the EFSA Comprehensive Database. The core atlas contains 3,990 market–food–contaminant profiles; the expanded explorer includes 9,248 derived food–analyte profiles. These are not 9,248 independent laboratory samples or distinct hazards. Processing notes identify recalculations, source mapping and unresolved verification steps, not official laboratory QA failures.

Serving estimates multiply mean laboratory concentrations by observed median portions among people who ate that food on the recall day. They are not average daily exposure in the population, habitual consumption, a prediction for a particular purchase or a dietary-risk assessment. For left-censored laboratory values, the lower calculation substitutes zero and the upper substitutes the reporting limit; neither asserts that the true level is zero or equals the limit. The displayed percentiles describe pooled or composite samples where applicable. Missing measurements do not imply zero contamination.

The Explorer offers downloads of its derived aggregate dataset, variable dictionary and current display/calculation JavaScript. These allow inspection of the displayed quantities, units and colour-bar formulae, but are not the original row-level laboratory records or the upstream cleaning and food-matching code. Full independent reproduction of that source-to-atlas transformation therefore remains an outstanding task.

The author’s background is in measurement and quantitative analysis, which informs the treatment of detection thresholds, grouping and uncertainty. This is not a substitute for formal toxicology expertise. Independent toxicology peer review has not yet been completed.

Additional context: EFSA on inorganic arsenic, EFSA on cadmium, FDA on lead and IRLs, and FDA/EPA advice on eating fish.