Soy has carried a contradictory reputation for twenty years. To one camp it is the most complete plant protein on the shelf; to the other it is an endocrine disruptor best avoided. Both sides cite “the studies”. The problem is that almost nobody reads them.

This article does. Every number below comes from a meta-analysis, a randomised controlled trial or a public health agency, and each source is linked directly so you can check it yourself.

The short version

• In adult men, the largest available meta-analysis (41 studies, up to 1,753 men depending on the measure) finds no effect of soy or isoflavones on total testosterone, free testosterone, oestradiol, oestrone or SHBG, at any dose or duration.
• In women, a meta-analysis of 47 trials finds no effect on oestradiol or SHBG. The only reproducible signal is a menstrual cycle lengthened by 1.05 days on average before menopause.
• On the thyroid, 18 randomised trials show no change in free T3 or free T4, and a very modest rise in TSH (+0.248 mIU/L) whose clinical meaning is undetermined.
• The genuine thyroid issue is not the hormone, it is drug absorption: soy can reduce the absorption of levothyroxine. You space out the tablet; you do not drop the soy.
• The LDL effect is real but small: roughly 4.8 mg/dL lower for 25 g of soy protein a day. Small enough that in 2017 the FDA proposed withdrawing its health claim for lack of consistency.
• About 77% of the world’s soy feeds livestock, and only around 7% goes to direct human food. Tofu is not what drives Amazon deforestation.

The rest of this article works through each of those points, including the limitations that summaries always drop.

Where soy comes from, and why it became contested

Soy (Glycine max) is a legume domesticated in East Asia. It has been cooked for a very long time in China, Japan, Korea and Vietnam as tofu, miso, natto and fermented sauces. You often read that it was one of China’s “five sacred grains”: that is a tradition recorded in old texts rather than a dated archaeological fact, and it should be presented as such.

What changed is the Western scale. Within a few decades soy moved from the Asian aisle to the protein aisle of every supermarket, carried by falling meat consumption. That visibility is what triggered the controversy. The more a food becomes a symbol, the more sweeping claims it attracts, in both directions.

What is actually in soy

These are measured values per 100 g from the US Department of Agriculture FoodData Central database.

Food (100 g)	Energy	Protein	Fat	of which saturated	Fibre	Calcium
Firm tofu, calcium sulfate set	144 kcal	17.3 g	8.7 g	1.3 g	2.3 g	683 mg
Tempeh	192 kcal	20.3 g	10.8 g	2.5 g	not reported	111 mg
Edamame, frozen, prepared	121 kcal	11.9 g	5.2 g	0.6 g	5.2 g	63 mg
Unsweetened fortified soy drink	33 kcal	2.9 g	1.6 g	0.2 g	0.5 g	124 mg

Three observations matter more than the table itself.

The calcium in tofu does not come from the bean, it comes from the coagulant. Tofu set with calcium sulfate carries 683 mg per 100 g; tofu set with nigari (magnesium chloride) carries far less. Two blocks sitting side by side on the shelf can differ fivefold on calcium. It is written in the ingredient list, and that is the only way to know. If calcium matters to you, the same logic applies across every non-dairy source.

Fibre is not a property of “soy” in the abstract but of the form you eat. Edamame give 5.2 g per 100 g, soy drink gives 0.5 g, because filtering removes the okara. Writing that “soy is high in fibre” about a glass of plant milk is simply wrong.

Soy is a complete protein source in terms of indispensable amino acids, which is uncommon among legumes. That does not make it compulsory: combining plant sources across the day gets you to the same place.

Isoflavones are not oestrogens

This is the heart of the misunderstanding. Soy isoflavones (genistein, daidzein, glycitein) have a structure that echoes oestradiol and can bind oestrogen receptors. From there came the shortcut: “soy contains female hormones”.

Two differences kill that shortcut. First, affinity: isoflavones bind preferentially to the beta receptor, whereas human oestradiol acts strongly on both, and their potency is orders of magnitude lower. Second, the effect depends on the hormonal background. Occupying a receptor without fully activating it can produce a mildly oestrogenic effect in one context and block a far more potent oestrogen in another.

The factor almost nobody mentions: equol

Daidzein can be converted by gut bacteria into S-equol, a considerably more active metabolite. Not everyone produces it. A review in Current Cardiology Reviews notes that only 20 to 30% of Westerners produce S-equol, against 50 to 70% of Asian populations (Sekikawa et al., 2019).

That figure explains much of the apparent chaos in the literature. Two people eating the same tofu do not generate the same metabolites. A trial run in the United States and a trial run in Japan are not working on the same metabolic population. When an Asian cohort reports a benefit an American trial cannot reproduce, it is not necessarily a contradiction; it may be two different microbiomes.

Does soy feminise men?

No, and this is the best documented question in the whole file.

The current reference is a meta-analysis published in Reproductive Toxicology in 2021, which extends and updates earlier work from 2010. It included 41 studies. Total testosterone was measured in 1,753 men, free testosterone in 752, oestradiol in 1,000, oestrone in 239 and SHBG in 967. The result is unambiguous: no significant effect on any of those measures, under any statistical model. Sub-analyses by isoflavone dose and by study duration change nothing (Reed et al., 2021).

The 2010 meta-analysis journalists usually quote, published in Fertility and Sterility, had already reached the same conclusion on a smaller body of evidence (Hamilton-Reeves et al., 2010). Fifteen further years of data did not make the feared effect appear.

So where do the gynaecomastia stories come from? From a handful of isolated case reports involving extreme intakes, on the order of several litres of soy drink a day for months. A case report describes what happened to one individual; it does not establish population risk. That is precisely the distinction headlines erase.

And in women?

The reference synthesis is a meta-analysis of 47 trials in Human Reproduction Update, covering pre-, peri- and post-menopausal women (Hooper et al., 2009).

Before menopause: no effect on oestradiol, oestrone or SHBG. FSH and LH fell by roughly 20%, and cycle length increased by 1.05 days on average (95% CI: 0.13 to 1.97 days). One day of cycle length is statistically measurable and practically invisible in a real life.

After menopause: no statistically significant effect on oestradiol, oestrone, SHBG, FSH or LH.

Hot flushes: a real but modest effect

A meta-analysis in JAMA pooled 62 trials and 6,653 women. Phytoestrogens were associated with a reduction of 1.31 hot flushes per day (95% CI: -2.02 to -0.61) and with an improved vaginal dryness score. Restricted to soy isoflavones alone, the effect drops to 0.79 flushes per day. Night sweats did not improve significantly. The authors flag that 74% of included trials carried a high risk of bias (Franco et al., 2016).

One fewer flush a day is not nothing when you get ten. It is also not hormone replacement therapy, and presenting it as such would be dishonest.

Thyroid: the real question is not the one people ask

The classic worry is that soy blocks the thyroid. The data do not support it.

A systematic review with meta-analysis in Scientific Reports gathered 18 randomised trials measuring free T3, free T4 and TSH. Results: no change in free T3 (mean difference 0.027 pmol/L, p = 0.499), no change in free T4 (-0.003 pmol/L, p = 0.656), and a rise in TSH of 0.248 mIU/L that barely reached significance (p = 0.049). The authors themselves conclude that the clinical significance of that rise, if any, is unclear (Otun et al., 2019).

The genuinely actionable point sits elsewhere, and it is missing from most consumer articles. A review in Thyroid examined 14 trials and reached two conclusions. In euthyroid people with adequate iodine intake, soy does not impair thyroid function. But in treated patients, soy can reduce levothyroxine absorption and therefore raise the dose required. The authors are explicit: hypothyroid adults need not avoid soy foods (Messina and Redmond, 2006).

In practice this looks like what we already do with calcium and iron: take the tablet away from the meal, keep a stable routine, and let the doctor adjust the dose on the basis of blood tests. The problem is not soy; the problem is changing your intake overnight without telling your doctor.

A theoretical reservation remains for people combining marginal iodine intake with already fragile thyroid function. That deserves a consultation, not a blog verdict.

Heart and cholesterol: a small effect, honestly stated

This is where many articles overreach.

The best available synthesis covers the 46 trials the FDA itself selected when it reopened its health claim. Across the 43 usable ones, a median dose of 25 g of soy protein a day over a median 6 weeks lowered LDL by 4.76 mg/dL (95% CI: -6.71 to -2.80) and total cholesterol by 6.41 mg/dL. The effect is statistically robust with no detected publication bias, but around a quarter of the trials showed no reduction at all (Blanco Mejia et al., 2019).

The regulatory context is telling. In October 2017 the FDA proposed revoking the “soy protein and coronary heart disease” health claim it had authorised in 1999, precisely because results lacked consistency from trial to trial (Federal Register, 2017).

What I take from this: the effect exists, it is real, and it is small. 25 g of soy protein a day is roughly 150 g of firm tofu, every day. For that, you gain a few milligrams of LDL. The main benefit of a tofu dish probably comes not from the soy itself but from whatever it displaced on the plate.

A dose-response meta-analysis of 23 prospective studies covering 330,826 participants separately found an inverse association between soy intake and cancer mortality (relative risk 0.88, 95% CI: 0.79 to 0.99). Association is not causation: people who eat tofu differ from people who do not along many other dimensions (Nachvak et al., 2019).

Breast cancer: the fear runs backwards

Soy was long discouraged for women who had had hormone-receptor-positive breast cancer. That advice came from rodent experiments, at doses far above human dietary intakes, in species that metabolise isoflavones differently from us.

Human data point the other way. The After Breast Cancer Pooling Project followed 9,514 survivors of invasive breast cancer in the United States and China for a mean 7.4 years, recording 1,171 deaths and 1,348 recurrences. Consuming at least 10 mg of isoflavones a day after diagnosis was associated with a statistically significant reduction in recurrence risk (hazard ratio 0.75, 95% CI: 0.61 to 0.92). The reductions in breast-cancer-specific mortality (0.83) and all-cause mortality (0.87) did not reach significance (Nechuta et al., 2012).

The American Cancer Society updated its public position in March 2025: soy foods are safe, the body of human evidence shows either reduced risk or no change, and the confusion traces back to animal studies. The same page draws a clear boundary: the ACS does not recommend soy supplements for cancer prevention, because they contain far higher isoflavone doses than foods and the evidence is lacking (American Cancer Society, 2025).

That food-versus-supplement distinction is the single most useful dividing line in the whole subject. It holds on the European side too: in 2015 EFSA concluded that at the doses found in supplements on the EU market, isoflavones showed no harmful effect on the mammary gland, uterus or thyroid of post-menopausal women. The agency immediately fenced that conclusion in: it applies only to the doses and durations studied, three months to just under a year, and no data allow an assessment of safety in peri-menopausal women or those with a cancer history (EFSA, 2015).

If a hormone-sensitive cancer is part of your history, that conversation belongs with your oncologist. What the literature says is simply that there is no reason to ban tofu as a precaution.

Soy allergy: the real risk, the one people forget

While everyone argues about isoflavones, there is a documented and immediate hazard: allergy.

Soy is a mandatory declared allergen in Switzerland. Annex 6 of the FDHA ordinance on food information lists it explicitly, and article 11 requires it to stand out in the ingredient list through typeface, style, colour or another suitable means (SR 817.022.16, Fedlex). It is not left to the manufacturer’s goodwill. The same substance sits on the EU’s list of 14 declarable allergens.

A food allergic reaction ranges from hives to anaphylaxis. The NHS lists the signs that require an immediate emergency call: sudden swelling of the lips, mouth, throat or tongue, difficulty breathing, a sense of choking, skin or lips turning blue, grey or pale, sudden confusion or fainting (NHS, food allergy). If you carry an adrenaline auto-injector, use it immediately, then call for an ambulance even if the person seems to improve.

Soy lecithin hides in an enormous number of processed products, from chocolate to industrial bread. For cooking around the constraint, see the allergen-free cooking guide and the soy-free protein alternatives.

How much to eat, concretely

The Swiss Federal Food Safety and Veterinary Office does not count “soy portions” but portions of protein sources. One daily portion of a protein-rich food equals 120 g of plain tofu, tempeh or seitan, or 30 to 40 g of soy granules, or 60 g of dry pulses, or 2 to 3 eggs, or 100 to 120 g of meat or fish. The recommendation is to vary sources across the week, with pulses at least once and meat at most two or three times (FSVO, adult nutrition).

Translated into a real week: tofu two or three times, lentils or chickpeas twice, tempeh once, eggs if you eat them. Nobody needs to count milligrams of isoflavones.

The cases that warrant a medical conversation rather than an article: levothyroxine treatment, a history of hormone-sensitive cancer, a diagnosed allergy, and infant feeding. On that last point, soy-based infant formula is a paediatric prescription rather than a shelf choice; it is not the same thing as a plant drink sold in a carton.

The environmental argument, turned around

This is the most useful reversal in the article, because it circulates backwards in both camps.

Yes, soy is a major driver of deforestation. No, it is not your tofu. According to figures compiled by Our World in Data, more than three quarters (77%) of the world’s soy is fed to livestock for meat and dairy production. Most of the rest goes to biofuels, industry and vegetable oil. Only about 7% goes directly to human food as tofu, soy milk, edamame or tempeh (Our World in Data, Soy).

On tropical deforestation itself, cattle ranching accounts for 41% of losses and oilseeds as a whole for 18%, with Indonesian palm oil dominating that second block (Our World in Data, Drivers of Deforestation). There is also a precedent for regulation that worked: Brazil’s Soy Moratorium, signed in 2006 as the world’s first voluntary zero-deforestation agreement, under which major traders committed to stop buying soy from Amazon land cleared after that date.

So the practical conclusion is not “avoid soy for the planet”. It is that soy’s impact runs overwhelmingly through meat, and eating the soy directly short-circuits that step. On traceability, Switzerland has built a domestic food-grade soy supply chain, which I unpack in the local tofu economy in Switzerland.

The forms of soy and what they actually change

Tofu is pressed curdled soy drink. Its firmness comes from pressing, its calcium from the coagulant. Firm tofu pan-fries and grills, silken tofu blends into creams. Avoiding a bland result is entirely a question of marinade and texture.

Tempeh is whole soybeans fermented by Rhizopus oligosporus and pressed into a cake. It gives more protein than tofu weight for weight, with a distinct nutty flavour. You often read that fermentation makes it “more digestible”: that is mechanistically plausible, since fermentation breaks down some oligosaccharides, but the clinical demonstration is thin. I present it as a reasonable hypothesis rather than a fact. You can also make it at home.

Miso is a fermented soy paste, often with added rice or barley. It is a salty condiment, and that is the point: you use it by the spoonful, not by the portion. Its job is umami, as covered in miso in Western cooking.

Natto is soy fermented by Bacillus subtilis, stringy and pungent, rich in vitamin K2. Worth flagging to anyone on a vitamin-K-antagonist anticoagulant: vitamin K interferes with that treatment, and it is a consultation topic.

Soy drink is a filtered extract. It loses the fibre, and unless fortified it supplies neither calcium nor B12. Check the fortification on the label, as with any B12 source.

Textured soy protein is an industrial ingredient that works well in stews. It is not bad in itself, but it usually arrives inside processed products where the salt and additives matter more than the soy.

To choose between these forms by dish, see the guide to plant-based substitutes, and sweet silken tofu if soy mostly reaches your table as dessert.

Seven mistakes I keep seeing

1. Treating a case report as evidence. A man drinking three litres of soy drink a day is not a sample.
2. Conflating food with supplements. Isoflavone capsules deliver doses unrelated to a block of tofu, and agencies assess them separately.
3. Dropping soy after a thyroid diagnosis. The issue is the timing of the tablet, not the food.
4. Assuming “fermented” automatically means better. Fermentation changes flavour and probably digestibility, but miso is still very salty.
5. Counting soy drink as a calcium source without reading the label. Unfortified, it supplies almost none.
6. Avoiding tofu for the Amazon. The problematic supply chain runs through meat.
7. Relying on soy alone. Lentils, chickpeas and beans bring different fibres and micronutrients. Plant iron in particular absorbs better in certain combinations.

Frequently asked questions

Does soy lower testosterone? No. Across 41 studies and up to 1,753 men depending on the measure, no significant effect was detected on total or free testosterone at any dose.

Can you eat soy during pregnancy? Soy foods are part of everyday eating in many countries, including during pregnancy. Isoflavone supplements are a different question: EFSA’s safety data cover post-menopausal women and do not extend to pregnancy. If in doubt, ask your midwife or doctor.

Is organic, non-GMO soy healthier? It answers legitimate farming and traceability preferences. On the health outcomes discussed here, no solid data show a difference between organic and conventional soy.

How many isoflavones are in a portion? Content varies with variety, processing and which part of the bean is used, which makes per-portion figures unreliable. The breast cancer survivor studies use a 10 mg per day threshold, reached through regular consumption of soy foods.

Does soy need special soaking or cooking? Dry soybeans are treated like other pulses: soak, then cook thoroughly. Processed products such as tofu, tempeh and miso are ready to use.

Is tofu a good iron source? Firm tofu supplies 2.66 mg per 100 g according to the USDA, but this is non-haem iron, which is less well absorbed. Vitamin C in the same meal improves that absorption substantially.

What I take away

Soy is a food, not a medicine and not a poison. The hormonal fears do not survive contact with the meta-analyses. The benefits are real but smaller than the marketing suggests.

The two points that genuinely deserve your attention are elsewhere, and they are concrete: spacing soy away from levothyroxine, and soy allergy, which is a documented risk and can be severe. The rest is ordinary dietary variety.

Sources

• Reed KE et al., Reproductive Toxicology, 2021: updated meta-analysis on male hormones
• Hamilton-Reeves JM et al., Fertility and Sterility, 2010: earlier meta-analysis
• Hooper L et al., Human Reproduction Update, 2009: circulating hormones in women
• Otun J et al., Scientific Reports, 2019: effect of soy on thyroid function
• Messina M, Redmond G, Thyroid, 2006: soy, thyroid and levothyroxine
• Blanco Mejia S et al., The Journal of Nutrition, 2019: meta-analysis of the 46 FDA-identified trials
• Federal Register, 2017: FDA proposal to revoke the health claim
• Nechuta SJ et al., American Journal of Clinical Nutrition, 2012: soy and survival after breast cancer
• Franco OH et al., JAMA, 2016: phytoestrogens and menopausal symptoms
• Nachvak SM et al., Journal of the Academy of Nutrition and Dietetics, 2019: soy and mortality
• Sekikawa A et al., Current Cardiology Reviews, 2019: S-equol, heart and brain
• EFSA, 2015: isoflavones in food supplements
• American Cancer Society, 2025: soy and cancer risk
• Our World in Data: Soy and Drivers of Deforestation
• FSVO: dietary recommendations for adults
• Fedlex: FDHA ordinance on food information, SR 817.022.16
• NHS: food allergy
• USDA: FoodData Central