Denmark tests around 20,000 samples of pork from finishing pigs every year, which is 0.1 percent of the animals slaughtered. The number that come back above the maximum residue limit is zero to two a year.
A 2025 review of antibiotic residues in milk carries a different figure. Prevalence runs from under 1 percent in some European countries to over 80 percent in parts of Africa. Same product, same chemistry, two orders of magnitude apart. What decides it is whether the chain of controls actually works.
Three links, not one
Most people picture a single check: someone takes a sample off a shelf somewhere. There are three links, and each one answers a different question.
| Link | Who does it | What it checks |
|---|---|---|
| Withdrawal period | the farm | how many days after treatment the milk or meat stays out of the supply |
| Screening on delivery | dairy, meat plant | whether this particular batch contains substances that inhibit bacterial growth |
| National monitoring | veterinary authority | whether the sector as a whole has a problem |
The first link handles volume: if the withdrawal period is respected, very little reaches the plant at all. The second catches the individual tanker. The third looks at the picture from above and was never designed to catch individual offenders. Mixing up those roles is what produces complaints along the lines of “they only test 0.1 percent”.
Where a withdrawal period comes from
A withdrawal period is a measured number, not a rounded guess. A Korean study from 2025 shows the mechanics with gentamicin. Nine Holstein cows were given the drug intramuscularly twice a day, milk was sampled every 12 hours for a week after the last dose, and the samples were run by liquid chromatography with tandem mass spectrometry.
Concentrations were above the 0.2 mg/kg limit 12 hours after the final dose and below it in every cow from 36 hours onward. The depletion data gave a withdrawal period of 40 to 48 hours. The regulatory withdrawal period set for that product is three days.
That gap is the point. Measured 40 to 48 hours, written down as 72. The margin is deliberate, because cows differ, dosing differs, and the error is meant to land on the safe side.
What the plant tests when the milk arrives
The screening test at milk reception runs on bacteria. A tube holds spores of Geobacillus stearothermophilus in a growth medium with an indicator. With no antibiotic present the spores germinate, the cells grow, and the medium turns from purple to yellow. With an antibiotic present growth is blocked and the tube stays purple. Purple is the positive result, and the batch is rejected.
One such system was validated under an AOAC protocol, and its detection limits came out at or below the EU maximum residue limit for most substances across six families: penicillins, cephalosporins, tetracyclines, sulfonamides, macrolides and aminoglycosides. The test sees named classes at the regulatory level rather than “something bad” in general.
The plant has its own reason to care, and it bites harder than any regulation. Residues inhibit starter cultures, so milk carrying them yields neither yogurt nor cheese. The 2025 review lists interference with dairy processing as a consequence of residues in its own right, alongside effects on the human gut microbiota.
National monitoring: the Danish 20,000
The Danish case is worth reading because someone costed it. Researchers from the industry council and the veterinary administration compared random sampling with risk-based sampling, where more samples go to herds with a higher chance of a violation. The risk indicator was chronic pleuritis recorded at slaughter, with a within-herd prevalence above 40 percent.
Risk-based sampling found more residue cases for the same money, and Danish abattoirs moved to it in January 2016. Note what changed: the number of samples stayed put, the address they were collected from did not.
Where the chain breaks
An Iranian meta-analysis pooled 52 studies from 2000 to 2022. Total prevalence of antibiotic residues in dairy products came out at 29 percent, with a confidence interval of 15 to 43 percent, and the highest contamination level belonged to tetracycline at 663 µg/l with a spread of 1540. The authors name the causes plainly: withdrawal periods ignored, overuse in animals, contaminated feed.
A Chinese systematic review screened 3131 citations and kept 46 surveillance studies published between 1988 and 2020 across 26 provinces, covering 8788 milk samples. Residue rates fluctuate but have largely declined in recent years. The substances found most often were sulfamethoxazole, chloramphenicol and trimethoprim.
One more number shows the direction of travel. Sales of veterinary antimicrobials across Europe fell 48 percent between 2011 and 2023. In Poland over the same years they rose 19.5 percent. A European average hides countries moving in opposite directions.
Resistance: the hard question
The everyday version is simple. Antibiotics go into animals, resistant bacteria end up in meat, someone eats it and picks up an infection that no longer responds to treatment. A Dutch group tried to measure that link.
A pooled analysis of 35 studies covered more than 27,000 samples, 5808 isolates of Escherichia coli producing extended-spectrum beta-lactamases, and 812 plasmids. The team then compared how similar the resistance gene profiles were across 22 reservoirs: people, animals, food, water, the environment.
The closest match was between farming communities and their own animals, broilers and pigs, at 0.8 to 0.9. Isolates from the general population matched clinical settings, surface and sewage water and wild birds at 0.7 to 0.8, and livestock or food reservoirs at only 0.3 to 0.6. The authors put the conclusion carefully: the analysis failed to demonstrate a close epidemiological linkage between livestock farms and people in the general population.
That is a narrower statement than “resistance does not travel through food”. The work covers specific bacteria in one country, and it confirms the risk for people who work on farms. What it does not support is the store-to-plate-to-untreatable-infection chain as popular writing draws it. The same mechanism runs in crop production too: streptomycin against bacterial plant diseases builds resistance under regular use, which our wiki entry on fungicides and bactericides covers.
An honest correction: mice, doses and missing data
There is an argument on the other side, and it concerns trace doses. A Taiwanese group fed mice tylosin at the theoretical maximum daily intake, a residue-level dose rather than a therapeutic one. On a high-fat diet those mice gained weight faster, developed insulin resistance and showed a shifted gut microbiota, and the phenotype transferred to germ-free mice along with the microbiota. Mice are mice and the finding does not carry to humans, but it does not wave away either.
What the data does not contain. There are no Ukrainian figures: international databases return nothing usable on domestic milk and meat, so borrowing Iranian or Danish percentages would be a substitution rather than an answer. There is no basis for believing that pasteurization or cooking removes residues, and the evidence points the other way, since residues spoil starter cultures at the processing stage. And a detected molecule is not proof of harm, because a limit is a level set with a margin rather than the threshold of poisoning, the same logic that governs nitrates in vegetables.
The practical reading. Milk and meat that pass through large processors clear at least two of the three checks, so the thin spot is anything sold around the plant. Food of animal origin from GM-fed animals, incidentally, carries no GMO label at all, an exemption Ukraine’s new law kept in place. Milk bought straight from a neighbor’s cow goes through neither reception screening nor state sampling, and everything rests on whether the withdrawal period was kept. It is the principle that runs through pesticides allowed in organic farming as well: the question is not whether a substance was used, but whether anyone checks the result.
Key points
- Residue control has three links: the withdrawal period on the farm, screening when a batch is delivered, and national monitoring. They do different jobs, and judging the system by one of them misses the other two.
- Withdrawal periods are measured. Gentamicin in milk cleared the limit at 36 hours and gave a calculated 40 to 48, and the regulatory figure was set at three days.
- The reception test detects inhibition of Geobacillus stearothermophilus, with detection limits validated at or below the EU limit for six major drug families.
- Denmark runs 20,000 pork samples a year and finds zero to two above the limit. Risk-based sampling catches more for the same budget.
- Where the chain fails the numbers change: 29 percent in Iranian dairy products across 52 studies, over 80 percent in parts of Africa.
- European sales of veterinary antimicrobials fell 48 percent between 2011 and 2023, while Polish sales rose 19.5 percent.
- A Dutch analysis of 27,000 samples found no close link between farms and the general population. It did confirm the risk to farmers themselves.
- No Ukrainian residue statistics are available in open databases, and that is the largest hole in the subject.
How to read exceedance percentages, and why “residues found” and “unsafe” are separate claims, is covered in our piece on the Dirty Dozen list. The wider picture sits in our food safety hub.
The processor incentive is clearest in its absence. Where milk goes straight to a customer, nothing downstream fails loudly: what the required raw milk panel measures.
