The largest error in mycotoxin analysis occurs before the laboratory begins.
A lot of grain is not a homogeneous mixture. A few kernels can contain a large portion of the toxin load, and those kernels are not neatly dispersed. Anyone taking a single incremental sample from a 500-tonne silo measures accurately what is in that incremental sample and virtually nothing about the lot.
Regulation (EU) 2023/2782 has applied since 1 April 2024 and replaces the old sampling rules. For cereals, oilseeds and their products, a fixed system applies: divide the lot into sublots, take the prescribed number of incremental samples per sublot and combine them into an aggregate sample. For lots of 50 tonnes and above, this means one hundred incremental samples and 10 kg, or 2.5 kg for small grains.
| Lot weight | Number of incremental samples | Aggregate sample |
|---|---|---|
| up to 0.5 tonnes | 5 | 1 kg |
| 0.5 – 1 tonne | 10 | 1 kg |
| 1 – 3 tonnes | 20 | 2 kg |
| 3 – 10 tonnes | 40 | 4 kg |
| 10 – 20 tonnes | 60 | 6 kg |
| 20 – 100 tonnes | 100 | 10 kg |
For small grains, a quarter of the aggregate weight applies. Above 500 tonnes, the regulation refers to a separate, large-scale sampling route.
The error remaining after mixing and dividing the aggregate sample is often greater than the analytical error of the test. In its technical documentation, Charm Sciences publishes an example in which a 1-pound sample yielded a spread of 1 to 46.9 ppb aflatoxin, whereas 10 pounds yielded 11.6 to 28.4 ppb. The same strip, the same operator, a different sample.
Milling is the second step that determines the outcome. For a grist of which only 60% passes through a 20-mesh sieve, the relative standard deviation between replicates increases to approximately 40%. At 99% passing through the sieve, it drops to approximately 4%. A finer grist is not a minor detail; it is the difference between an actionable and an unusable result.
The regulation is clear on the decision rule. A lot is rejected if the analytical result, corrected for recovery and minus the expanded measurement uncertainty, exceeds the maximum level. A result falling within the uncertainty band of the standard is therefore not an automatic rejection — but neither is it proof of compliance. In that case, a second sample or confirmatory analysis is the only defensible route.
<a href="/en/kennisbank">Go to the knowledge base</a>
<a href="/en/normen">Go to the limits</a>