The Verification Venue · a famous scare, put on the stand
The Syndrome That Couldn't Pass a Blind Test
The "MSG headache" has a birthday: a single 1968 letter that only guessed. Since then, dozens of double-blind trials have tested it. Split them by how the MSG was given and the answer stops being a coin-flip: with food, nobody reacts. The only "yes" comes from a big dose dissolved in a drink on an empty stomach, and there the trials could not keep a secret their subjects could taste.
double-blind studies that gave MSG WITH FOOD and found more headaches than placebo (Obayashi & Nagamura 2016). The "yes" signals all came from a different experiment: 4 of 7 studies that dissolved a large dose in a drink, on an empty stomach, above the concentration where MSG's own taste gives the game away.
Robert Ho Man Kwok's 1968 letter to the New England Journal of Medicine coined "Chinese-Restaurant Syndrome" and floated three suspects for the numbness and palpitations he felt after dinner: the cooking wine, the high sodium, and the monosodium glutamate. MSG was one guess of three, on one man's anecdote. It is the one that stuck, got a racialized name, and outran its evidence for fifty years. Below, you can run the actual evidence: pick a dose, pick whether it rides in on food or in a fasting drink, and watch what the pooled trials actually found, plus a live flag for whether the study could even stay blind.
From a heavily seasoned restaurant portion (~0.5 g of added MSG, eaten across a meal) up to the biggest experimental bolus (9 g in one drink). The FASEB report's threshold for even transient symptoms in a sensitive person is 3 g taken without food, far above any meal.
"In food" is how anyone actually eats MSG. "Fasting, in a drink" is the lab exposure: the dose dissolved in a fixed 150 mL serving, on an empty stomach.
Pooled double-blind verdict, this regime
–
–
Blinding intact?
–
–
Vs a real meal
–
–
Each dot is a real double-blind study, placed by its MSG dose. Filled = it found more headache than placebo; hollow = no difference. Click any dot (or any row in the table below) to see its numbers. Note where the filled dots live: only on the fasting-drink lane, and only to the right of the taste line, where the drink is strong enough that a subject could tell they got the real thing.
Click a study to inspect it
The twist is context, not chemistry
The verdict flips on HOW it was given, and the "yes" arm is an artifact
1. With food, the effect is a clean null. Six double-blind studies gave MSG the way people eat it, mixed into a meal, at doses up to 3 g. Not one found more headache than placebo. (One 1973 study saw a signal in a female-only subgroup: 6 headaches on MSG vs 1 on placebo. A single subgroup in a single study, and the overall result was still null.) This is the regime that matches dinner, and it says nothing is there.
2. The only positives are broken-blinding artifacts. Every "yes" came from dissolving a large dose in a drink on an empty stomach. But MSG has a distinct savory taste that is detectable above about 1.3 % (2 g in 150 mL). Every positive drink dose (2.5 g and up) sat above that line, so subjects could taste which cup was "real." A trial you can taste your way through is not truly blind. Re-blind it, by putting the MSG in food or using a dietary dose, and the signal disappears.
So the honest claim is not "MSG is inert." It is: no proven causal link to headache at the doses and contexts you actually consume; a small sensitive subset may react to a big fasting dose (the FDA and a 1995 FASEB report allow this for at least 3 g without food); and the studies that said otherwise could not keep their subjects blind. The review's own words: "a causal relationship between MSG and headache has not been proven." A null result, not a clean bill of health.
The question the folk story asks
"Does the MSG in my dinner give me a headache?" Tested directly (MSG with food, dietary doses): 0 of 6 studies found an effect. Answer: no reliable effect.
The question the scary studies answered
"Can a big fasting bolus of MSG in a drink you can taste produce more reported headaches?" Sometimes (4 of 7) yes, but the taste leaked the assignment, and nobody eats this way. Different exposure, broken blind.
And the cultural edge, stated plainly: the "Chinese Restaurant Syndrome" label singled out one cuisine for a molecule that is naturally abundant in tomatoes, aged parmesan, and human breast milk. The scare was pointed at a culture before it was pointed at a dose. Even the official headache classification (ICHD) once listed MSG as a trigger and has since dropped it, which is exactly why a search still finds the "yes" answer in respectable places. The evidence moved; the folklore, and some of the SERP, did not.
The thirteen studies, digitized
This is the table the review built, transcribed. Rows are clickable. The two with published both-arm rates draw real bars above; the rest report a dose and a significance verdict. Read the pattern: the "with food" block is null top to bottom; the "without food" block is where the positives are, and every one of them is at a taste-detectable concentration.
| study | dose (g) | MSG % | placebo % | result | blind? |
|---|
The check: every number recomputed in front of you
Nothing here is a smooth fabricated curve: the dots are the real trials. What the page computes live is the bookkeeping, the taste/blinding threshold for your dose, the with-food and without-food positive counts, and the Geha vs placebo gap:
The offline gate recomputes all of this from the digitized table: node research/does-msg-cause-headaches/verify-does-msg-cause-headaches.mjs. Free choices & uncertainty: the taste/blinding threshold (1.3 % = 2 g / 150 mL) is the review's own figure and depends on serving volume and individual sensitivity; the concentration flag uses a fixed 150 mL serving (Geha's basis). Doses given as ranges (Prawirohardjono 1.5–3 g, Rosenblum 5–12 g) are plotted at a representative value but their result verdicts are exact. "Positive" means the study's own significance test, not a re-analysis. The dose-response is deliberately not drawn as a curve: the data are inconsistent (Baad-Hansen was positive at 4.5 g but null at 9 g; Rosenblum null up to 12 g), which is itself the honest reading behind "not proven."
What's exactly true here, and what's a judgement call
Exactly true (the counts and the quotes). The best systematic review of human studies (Obayashi & Nagamura 2016) sorts the double-blind trials by context: of 5 papers / 6 studies that gave MSG with food, none found more headache than placebo (a single female subgroup in Zanda 1973 aside); of 5 papers / 7 studies that gave MSG without food in a drink, 4 found a signal. Its verdict is that a causal relationship "has not been proven." Geha et al. 2000 reported headache in 54 % on 5 g MSG vs 28 % on placebo (P < 0.005), in an unblinded-by-taste fasting drink. MSG is detectable at 1.3 % (2 g / 150 mL). Kwok's 1968 letter named MSG as one of three speculative causes.
The honest limits. "Not proven" is a null, not a proof of safety: the FDA holds MSG generally recognized as safe (GRAS since 1959), and the 1995 FASEB report allowed that a small, sensitive subset may get transient, mild, short-lived symptoms after at least 3 g of MSG taken without food. That is a real caveat, and it is still a dose no seasoned meal delivers. We do not claim MSG causes nothing; we claim there is no reliable headache effect at dietary doses in food, and that the positive drink studies could not stay blind.
A judgement call (the blinding critique). The taste-threshold argument is the review's, and it is a critique of internal validity, not a re-run of the trials. The positive studies are real; the point is that their design let the assignment leak through the tongue, which is why the effect vanishes when the MSG is hidden in food. Reasonable people can weigh that differently; the review weighs it toward "not proven."
Two things we deliberately do not say. We do not state that Kwok's letter was a deliberate hoax: that story (a claimed prank) is disputed and unverified. And we do not treat the systematic review as "proving MSG safe": it concludes the evidence is inconsistent and the question is unresolved by a null, which is precisely the honest position.