How much milk and peanut turns up in chocolate that does not list them, what a 'may contain' label does and does not tell you, and how hard it is to clean one chocolate out of a line before making another.
8 studies2003 to 2026Each checked against its published abstract
These are studies, not advice. Nobody should decide what is safe for them to eat from a survey of other people’s chocolate. For what the law requires on a label, see allergen labelling rules.
What these studies add up to
Milk in dark chocolate is the best-documented case. In a survey of bars sold in the United States, milk was measurable in three-quarters of the bars that carried a milk warning and in a third of the bars whose labels did not mention milk at all, often at more than 1,000 parts per million.
The wording of a warning told the buyer nothing about the amount. Neither the type of advisory statement nor its absence predicted how much milk a bar held, and a British study of other foods reached the same conclusion about precautionary labels generally.
The reason is mechanical. Chocolate lines are cleaned dry, because water is a hazard of its own in a chocolate factory. In pilot trials, pushing dark chocolate through equipment that had run milk chocolate took kilograms of product to bring milk below the limit of the test.
The doses that matter are small. Modelling of supervised food challenges put the dose at which one allergic person in a hundred reacts at a tenth of a milligram of protein for milk and two tenths for peanut.
Testing is not a simple check either: tempering reduced how much allergen an immunoassay could detect in a model chocolate.
Most countries regulate allergens used as ingredients. Very few regulate traces, and an international expert panel has agreed what a precautionary label ought to mean without any law yet requiring it.
Peanut protein in chocolate bars that did not list peanuts
Do chocolate bars that do not list peanuts contain peanut protein?
How
92 chocolate bars bought in Canada, 32 made in North America and 60 imported from Europe, tested with one commercial immunoassay.
Finding
None of the 32 North American bars held detectable peanut protein. Among bars with no precautionary label, 30.8% of those from western Europe and 62% of those from eastern Europe did, at up to 245 parts per million.
Does not establish: Anything about bars on sale today, in Europe or elsewhere.
Limits: One assay, one purchase, over twenty years ago, before the present European labelling law.
Below what amount of an allergen do nearly all allergic people not react?
How
Statistical modelling of individual thresholds from more than 55 studies of supervised oral food challenges, for 11 foods.
Finding
The dose expected to bring on a reaction in 1% of the allergic population was 0.2 mg of protein for peanut, 0.1 mg for cow's milk, 0.03 mg for egg and 0.1 mg for hazelnut.
Does not establish: A safe amount for any individual. These are population estimates with 1% reacting by definition.
Limits: People who join challenge studies may not represent everyone with the allergy, and the most severely affected are often excluded.
Does chocolate processing change how well tests recover peanut, egg and milk allergens?
How
A model dark chocolate with known amounts of peanut, egg and milk added, extracted with five different buffers and examined with antibodies and with serum from allergic people.
Finding
Tempering decreased antibody binding whichever extraction was used. Detergent or a reducing agent improved recovery of peanut proteins but not of egg or milk.
Does not establish: That tempered chocolate is less allergenic. Less was detected; nothing shows less was there.
Limits: A model chocolate made for the purpose. The abstract gives no figures for the size of the loss.
What risk do unintended allergens pose in British foods, and does a precautionary label mark the riskier ones?
How
Measured allergen concentrations in UK products across several categories, with and without precautionary labels, fed into a probabilistic model of reactions in people allergic to milk, wheat, hazelnut and peanut.
Finding
Most products that tested positive held enough allergen to be predicted to cause a reaction in more than 1% of the allergic population. That was true of products with a precautionary label and of products without one.
Does not establish: That labels are worthless. It shows the label and the measured risk were not well matched in this selection.
Limits: The abstract does not give the number of products, and the result applies to those that tested positive, not to all those sampled.
How often do dark chocolate bars sold in the United States contain milk or peanut that is not an ingredient?
How
88 dark chocolate products tested for milk and 78 for peanut, three samples of each, with two immunoassay kits per allergen.
Finding
Milk was measurable in 75% of bars carrying a milk advisory statement, mostly above 1,000 parts per million; in 33% of bars whose label did not mention milk; in 25% of those labelled vegan; and in 15% of those labelled dairy free or lactose free. Peanut was found in 8% of bars with an advisory statement and 17% of bars with none.
Does not establish: The rate today, or outside the United States. It does establish that neither the wording of a warning nor its absence predicted how much milk a bar held.
Limits: The authors call it a limited survey. It is one country, one period, and not a random sample of the market.
Where does precautionary allergen labelling stand worldwide, after an expert consultation convened by the FAO and WHO?
How
A review of the issues, described by its authors as non-systematic.
Finding
Approximately 100 countries have legislation on declaring allergenic ingredients. Just a few have legislation on allergens present unintentionally, and precautionary labels are applied inconsistently.
Does not establish: Which countries those few are, or what their rules require. The abstract does not list them.
Limits: Not a systematic review, and written by a body that advocates regulation.
How well do dry cleaning methods remove milk chocolate from equipment before dark chocolate is made on it?
How
Pilot-scale trials in triplicate on valve and pipe assemblies, a ball mill, a conche and a roll refiner, comparing no cleaning, a pipeline pig, a cocoa butter flush and wet cleaning, with milk measured by immunoassay.
Finding
With push-through alone, the first dark chocolate out of a valve held 4,500 micrograms of milk per gram or more, and at least 16.2 kg had to pass before milk fell below the assay's limit of 2.5. Without cleaning, first batches from a ball mill and conche held 17,000 micrograms per gram or more; a cocoa butter rinse cut that by at least 89%.
Does not establish: A flush quantity that would work on any other line. The paper shows the scale of the problem, not a procedure.
Limits: Pilot equipment and one recipe. Factory lines are longer and have more places for residue to sit.
What do clinicians, patient representatives and others agree precautionary allergen labelling should be?
How
A Delphi study: 35 statements put to a panel of 35 experts over 3 rounds, with consensus defined as 70% agreement.
Finding
Consensus was reached on 33 of 35 statements. The panel agreed that a precautionary label should be used only when an unintended allergen may be present above a set action level, should rest on a formal risk assessment, and should not appear beside a 'free from' claim for the same allergen. There was no consensus on how to tell consumers about risk below the action level.
Does not establish: That any label on sale follows these principles.
Limits: A consensus of invited experts. It records agreement, not evidence, and binds nobody.
How much unintended milk and peanut is in chocolate on sale now. The surveys read here are from 2003 and 2017, and each is one country at one time.
How common tree nut, soy, wheat and sesame traces are in chocolate. No survey of any of them in chocolate was read.
Whether reference doses will be written into law, and at what level. The expert panel preferred a dose protecting 95% of allergic people to one protecting 99%, to limit over-labelling.
What dry cleaning can reliably achieve on a full-scale line, as opposed to pilot equipment.
These are questions the studies read for this page leave unanswered. They are not a claim that nobody has answered them.
We would like to set two Google Analytics cookies to count page views, and to count them in our own cookie-free measurement as well. They do not identify you and nothing is shared with advertisers. Decline and the site works exactly the same — see cookies for what each one does.