If you’ve spent any time on migraine forums, you’ve seen the same list a hundred times: chocolate, aged cheese, red wine, cured meats, citrus, MSG. The list has been around for decades. It is also mostly built on weak evidence and recall bias, and the one prospective cohort that tested it carefully found that food triggers in general don’t predict attacks usefully.

This page is the corrective. If you want to skip to the practical implication: don’t restrict your diet based on a list you found online. Talk to a clinician, and consider the bidirectional direction-of-causation problem before concluding a food caused your attack.

The famous list and where it came from

The canonical “top migraine food triggers” list — chocolate, aged cheese, red wine, cured meats, citrus fruits, MSG, artificial sweeteners — traces back to a handful of small studies from the 1970s and 1980s, plus a long tradition of patient self-report. The biochemical hypotheses (tyramine in aged cheese, sulphites in wine, nitrates in cured meats, phenylalanine in chocolate) sound mechanistic but have never been replicated in well-controlled provocation studies at the population level.

The 2018 Pellegrino et al. meta-analysis in Cephalalgia reviewed perceived-trigger research across multiple cohorts. Self-reported triggers were prevalent but rarely confirmed by controlled provocation. Many show what researchers call a “perceived-but-not-proven” pattern — patients believe the trigger, the data does not.

The Holsteen 2020 cohort that tested everything at once

The most rigorous test of food triggers in a recent prospective cohort was Holsteen et al., 2020 in Headache. They followed 178 patients for up to 90 days each, with daily logging of caffeine, alcohol, sleep, stress, menstruation, self-prediction, and a comprehensive food checklist. They then trained a within-person model on the resulting data.

Result: within-person C-statistic of 0.56. For context, 0.50 is no better than chance. The model couldn’t distinguish attack days from non-attack days using self-reported trigger exposure.

The paper’s verbatim conclusion: “Episodic migraine attacks were not predictable based on self-prediction or on self-reported exposure to common trigger factors.” That includes food.

The recall-bias and direction-of-causation traps

Two specific issues haunt food-trigger self-report:

  1. Recall bias. You’re far more likely to remember what you ate the day before a migraine than the day before an uneventful day. Statistically this manufactures associations that aren’t real. The only fix is prospective daily logging whether or not you have an attack — and even then, the second issue intervenes.
  2. Direction of causation. Migraine prodromal symptoms include appetite changes and food cravings — particularly for salt, chocolate, and carbohydrates. So “I ate chocolate yesterday and had a migraine today” is consistent with both directions: chocolate caused it, or the prodrome caused the chocolate craving. From self-report alone, you cannot tell.

This second one matters a lot. A person who craves chocolate during prodrome may correctly observe a strong correlation between chocolate and migraine — and still be drawing the wrong causal conclusion.

When food triggers are real

For a small subset of people, specific foods do appear to trigger attacks reliably, in a way that survives controlled provocation and personal n-of-1 testing. The honest signs you might be in that group:

  • Reproducibility under fasting state — the food triggers reliably even when you weren’t craving it.
  • Consistent timing — onset within a narrow, repeatable window (e.g., always 6–8 hours after exposure).
  • Survives a controlled rechallenge — after eliminating for several weeks and reintroducing, the trigger fires again.

If those three are present, your direct experience is probably informative. If they aren’t, you may be in the larger group where the perceived trigger is artefact.

Why an elimination diet is the wrong default

The temptation to fix migraine by eliminating foods is understandable. The cost is real:

  • Nutritional risk — broad eliminations (gluten, dairy, all tyramine sources) can produce deficiencies if maintained.
  • Social cost — restrictive eating affects family meals, travel, work events.
  • Opportunity cost — time and attention spent on diet aren’t spent on the interventions that have stronger evidence (sleep, preventive medication, weather planning).
  • Doesn’t generalise — even if your trigger is real, the family member or friend who shares your “list” may have entirely different triggers or none at all.

A registered dietitian who specialises in headache nutrition can run a structured elimination protocol that minimises these costs. A printed list from a wellness blog cannot.

Why Hermly doesn’t track food

We considered adding food logging. We chose not to because:

  1. The Holsteen 2020 evidence is clear that food self-report doesn’t meaningfully predict.
  2. Daily food logging is high-friction. Even if it added 1–2 AUC points, the retention cost would dominate.
  3. Including a feature implies it’s useful. Showing a “Food” factor card would tell users we think food predicts their attacks — which we don’t.

The 26 features Hermly does read are the ones with prospective evidence behind them. Food, despite being on every popular trigger list, is not in that set.

What this isn’t

Not a claim that food never triggers migraines. For some individuals, a specific food consistently causes attacks — and for them, identifying and avoiding it is the right move. What this is arguing against is the population-level assumption that chocolate, cheese, and red wine are common triggers for most people. The evidence doesn’t support that, and acting on it as if it does has real costs.