Mycotoxin testing: why spot sampling fails

The uncomfortable truth is that the number may tell you very little about the lot you actually received.

If the sampling plan consisted of one grab from the top of a silo or a quick scoop from a moving belt, you did not test the commodity in any meaningful statistical sense. You tested the few grams that happened to enter the container. The lot may weigh several tonnes or several hundred, and the contaminated material may be concentrated in a small, unevenly distributed pocket.

This is the central problem behind mycotoxin testing sampling plan errors. The laboratory can perform the extraction perfectly and report a precise result. If the submitted sample did not represent the lot, the precision applies only to the wrong material.

It is not primarily a problem with the HPLC, the ELISA reader, or the analyst. It is a problem with the sample.

The Physics of Heterogeneity: Why Mycotoxins Cluster in Hot Spots

Mycotoxins are not mixed into grain like sugar into cake batter. They do not diffuse evenly through a bulk lot, and they do not politely average themselves out before inspection. They concentrate where the fungus grew.

A single Aspergillus flavus infection in maize can produce very different toxin levels from one ear to the next, even when the plants grew in the same field. Drought stress may affect one part of a field more severely than another. Insect damage creates wounds that favor mold colonization in particular plants. During storage, moisture can migrate and condense against walls, beneath roofs, around damaged grain, or in other poorly ventilated areas. One section of a bin may remain relatively dry while another develops the conditions needed for fungal growth.

These processes create what the industry calls hot spots: localized pockets of contamination that may be far more concentrated than the surrounding commodity.

The problem is not limited to grain. Similar patterns occur in nuts, dried fruit, oilseeds, spices, and compound feed. The physical units are different, but the sampling challenge is the same. A few affected kernels, pods, fruits, or feed particles can carry a disproportionate share of the total toxin burden.

The contaminated portion may represent only a small fraction of the total mass and still influence the safety decision. A 100-tonne bulk lot with a 200 kg contaminated pocket is not uniformly contaminated. But if that pocket enters the product stream, it can dominate the result for the material made from it. If a grab sample misses the pocket, the laboratory may report a low concentration while the next portion of the lot presents a very different exposure profile.

This is why visual inspection is a weak substitute for representative sampling. Moldy kernels may be obvious in a heavily affected area, but mycotoxins themselves are invisible. The fungus may no longer be active. The commodity may look and smell acceptable while the toxins produced earlier remain stable in the material.

A sorting line can remove visibly damaged units. It cannot prove that an apparently clean bulk lot is toxin-free. The absence of visible mold is not evidence that a sample is representative, and a clean-looking sample is not evidence that the lot is clean.

Why a single grab is structurally weak

A single grab has several predictable weaknesses:

• It usually samples one accessible location rather than the entire lot.
• It overrepresents the surface, the top of a truck, or the easiest section of a bin.
• It cannot distinguish a genuinely clean lot from a contaminated lot in which the hot spot was missed.
• It gives no information about how contamination changes along a conveyor, discharge stream, or storage structure.
• It makes the result highly sensitive to the physical location and timing of collection.

The problem is not that the individual grab is chemically invalid. The problem is that it carries too much responsibility. One small portion is being asked to stand in for a heterogeneous mass that may have been formed through different field, transport, drying, and storage conditions.

Mycotoxins do not blend. They cluster. If your sampling plan does not reflect that, your test result is decoration, not data.

Quantifying the Failure: How Sampling Error Outweighs Lab Analysis

The most important distinction in a mycotoxin result is the difference between analytical error and sampling error.

Analytical error arises after the laboratory receives the material. It includes extraction variability, instrument performance, calibration, recovery, repeatability, and the uncertainty associated with the method. Sampling error occurs before that: it is the difference between the material submitted to the laboratory and the lot from which it was taken.

In heterogeneous commodities, sampling error can dominate the total uncertainty. Technical literature has often described sampling as responsible for more than 90% of the total variance in mycotoxin testing under difficult bulk-commodity conditions. The exact proportion depends on the commodity, toxin, lot, sampling design, sample size, and analytical method, so it should not be treated as a universal constant. The underlying lesson is robust, however: improving laboratory precision does not compensate for an unrepresentative sample.

Modern laboratories may deliver highly repeatable measurements on the aliquot placed in the instrument. A method can produce a low coefficient of variation and still fail to describe the lot if the aggregate sample was collected from the wrong places.

That is the uncomfortable asymmetry:

• The laboratory may measure a few grams with high precision.
• The few grams may have come from a kilogram-scale composite.
• The composite may have been drawn from a single point in a tonne-scale lot.
• The reported result may then be used to make a decision about the entire shipment.

At each step, a smaller portion is used to represent a larger one. The process works only if the material has been collected, mixed, reduced, and prepared in a way that preserves representativeness.

Why more laboratory precision cannot rescue poor sampling

Suppose an analyst receives a finely homogenized portion and runs it twice. The two readings may agree closely. That tells you the instrument and the method are behaving consistently for that portion.

It does not tell you whether another portion of the same shipment would produce the same result.

If the original sample missed a hot spot, repeating the analysis confirms the concentration in the clean material. If the original sample captured an unusually concentrated pocket, repeating the analysis confirms that pocket. Replicating the laboratory measurement reduces analytical uncertainty; it does not remove the uncertainty created when the bulk sample was collected.

This is also why spot sampling can fail in both directions. It can fail to reject a contaminated lot because the sample misses the affected material. It can also overstate the condition of a broadly clean lot if one small, unusually concentrated fragment enters the sample. Neither outcome is necessarily a laboratory mistake. Both are consequences of asking an inadequate sampling design to answer a lot-level question.

A composite sample is not automatically representative, either. Combining several poorly chosen grabs does not fix a biased collection process. Ten scoops from the same convenient point are still a convenient-point sample. The increments must be distributed across the relevant volume or time period, and the aggregate must be large enough to preserve the chance of capturing uneven contamination.

The industry shorthand is that the sample is the test. That is not a slogan about laboratory technique. It is a reminder that the laboratory cannot recover information that was never placed in the sample container.

Most of the risk is upstream of the laboratory. A better instrument cannot correct a sample that never represented the lot.

Regulatory Benchmarks for Representative Aggregate Sampling

Once sampling is recognized as the main source of uncertainty, the regulatory logic becomes easier to understand. Official methods do not rely on one convenient scoop. They specify how a lot is identified, how increments are distributed, how much material is combined, and how the aggregate is reduced for analysis.

The European framework has also changed, and the distinction matters. Commission Regulation (EC) No 401/2006 was the long-standing reference for methods of sampling and analysis for the official control of mycotoxins in foodstuffs. It has since been superseded for this purpose by Commission Implementing Regulation (EU) 2023/2782, which applies from 1 April 2024. Older technical materials may still cite Regulation 401/2006, but current procedures should be checked against the requirements in Regulation 2023/2782 and any applicable national implementation or official guidance.

The basic principle remains familiar: the larger and more heterogeneous the lot, the more carefully the sampling plan must distribute incremental samples and define the aggregate sample.

A lot is an identifiable quantity of a commodity, delivered or produced under conditions presumed to be uniform with respect to characteristics such as origin, variety, packaging, packer, consignor, marking, and other relevant identifiers. A truckload, bin load, or barge load can constitute a lot when it is an identifiable quantity with common characteristics. The container does not determine the lot by itself, and one physical container can hold more than one lot if the material is not uniform or is separately identified.

That definition is more useful than a categorical rule because it connects the sampling unit to the decision being made. The question is not whether the material happens to be in a truck or a silo. The question is whether it can reasonably be treated as one identifiable, comparable quantity.

Aggregate samples and incremental sampling

Under the applicable EU sampling schedules, requirements vary by commodity and lot size. The numbers below illustrate the scale of the task rather than replacing the current regulation for every product category.

For certain dried-fruit lots in the 15–30 tonne range, the prescribed approach uses an aggregate sample of 30 kg assembled from 100 incremental samples. For a large cereal lot of around 500 tonnes, the relevant schedule may require an aggregate sample of at least 10 kg, also assembled from 100 increments.

The precise requirements depend on the commodity, lot size, presentation, and regulation in force. The point is not that every lot requires the same fixed combination. The point is that official control sampling is designed around many increments and a substantial aggregate, not a single visual check.

The incremental count improves coverage. If contamination is patchy, each additional increment gives the sampling plan another opportunity to encounter affected material in a different part of the lot. It does not make the result magically exact, and it cannot compensate for a badly defined lot or a poorly mixed aggregate. It does, however, reduce the dependence on one location.

The aggregate weight serves a different function. Once increments have been collected, enough material must remain to mix, grind, divide, and submit a valid laboratory portion. Reducing the sample too early can undo the value of the collection stage. A small shipment sample may be easier to move, but convenience is not a valid reason to discard material before representative reduction.

Where the increments come from

The pattern matters as much as the count.

For material moving through a conveyor or discharge stream, increments should be collected at defined time intervals throughout the relevant flow. For a static lot, the plan may require probing different locations and depths. For packaged commodities, selected units must be distributed through the lot rather than taken from the first accessible cases.

The sampling method should account for the way the lot is physically handled. A probe that reaches only the top layer cannot represent material settled at the bottom. A scoop taken only at the beginning of discharge cannot represent what emerges later. A sample from the front of a truck may reflect loading order rather than the shipment as a whole.

Common shortcuts include:

• taking all increments from the top of a truck or bin;
• collecting only from the discharge point that is easiest to reach;
• stopping once the container looks full;
• choosing intervals based on operator convenience rather than the written plan;
• mixing different origins or storage units into one nominal lot;
• reducing the aggregate before it reaches the laboratory without a validated procedure;
• recording the final result without recording how the sample was obtained.

These shortcuts are attractive because they save time. They also make the certificate difficult to interpret when the result is challenged.

Outside the European Union, the legal requirements and terminology may differ. U.S. Federal Grain Inspection Service procedures, for example, operate under a different regulatory structure. The technical logic is nevertheless similar: defined lots, systematic incremental sampling, controlled aggregation, and documented sample handling are essential when a result carries commercial or regulatory weight.

The result is only as defensible as the record

A useful certificate should connect the analytical number to the material and procedure behind it. The record should identify the lot, sampling location, date, commodity, quantity, sampling equipment, number and distribution of increments, aggregate weight, reduction method, and chain of custody.

This documentation does not make an unrepresentative sample representative after the fact. It does something else: it shows whether the result can be interpreted at all. Without that information, a low number may look reassuring while concealing a sampling process that could never have detected a localized hot spot.

The Hidden Challenge of Masked Mycotoxins in Feed Analysis

Even a well-designed sampling plan leaves a second analytical question: what exactly did the method measure?

Some plants modify mycotoxins through their own metabolism. The resulting conjugated or modified forms are often called masked mycotoxins because a routine method aimed at the parent compound may not detect them directly. A common example is deoxynivalenol-3-glucoside, a plant-associated conjugated form of DON.

The term masked can be misleading if it suggests that the toxin has disappeared. The molecule has changed, and its behavior may change with it, but some modified forms can be transformed during digestion and contribute to exposure. The toxicological importance depends on the compound, the animal or person exposed, the matrix, the extent of conversion, and the available evidence.

Routine testing panels are often built around parent compounds such as aflatoxin B1, deoxynivalenol, zearalenone, ochratoxin A, or fumonisins. A method calibrated only for those targets may not report every conjugated or bound form in the sample. The instrument can be working exactly as intended while the reported result describes only the compounds included in the method.

The size of the gap is not constant. Published research has reported that masked forms can represent a substantial share of the total mycotoxin burden in certain commodities and under particular conditions, with some studies estimating contributions as high as 60% for specific combinations of toxin and matrix. That figure should not be generalized to every feed or grain lot. It is a warning about scope: a parent-toxin result is not automatically a total-toxin result.

Direct measurement versus potential exposure

Laboratories address this problem in different ways.

One approach is a broader LC-MS/MS panel that measures selected modified forms alongside the parent compounds. This gives separate results for the analytes included in the method, but it still depends on the laboratory’s target list and validation.

Another approach uses enzymatic or simulated gastrointestinal digestion to examine how conjugated forms may be converted into detectable parent compounds. Such procedures can provide an estimate of total or potentially bioavailable toxin, depending on the design of the test. They do not reproduce every detail of digestion, and the result should not be confused with a universally accepted measure of in vivo exposure.

For feed analysis, the right question is therefore not simply whether the laboratory tests for mycotoxins. It is which mycotoxins, which modified forms, which matrix, and which analytical endpoint.

Before comparing two certificates, check whether they describe the same thing:

• Is the result for the parent compound only?
• Are selected masked forms included separately?
• Is the reported value a sum of defined analytes?
• Was a digestion or deconjugation step used?
• Are the limits of detection and quantification suitable for the intended decision?
• Does the method apply to the actual feed matrix rather than an easier reference material?

A compliance decision based on a narrow panel may be entirely appropriate when the applicable rule defines that panel. It should not be presented as a complete assessment of dietary or animal exposure without further qualification.

If a method only sees what it was designed to see, it has measured its target list—not necessarily the full risk in the material.

Standardizing Particle Size: The Role of Milling and Sieving

When the aggregate sample reaches the laboratory, the sampling problem changes scale. The material may be representative at the lot level, yet still be too heterogeneous for the small test portion placed in the extraction vessel.

That is why grinding and homogenization are not cosmetic preparation steps. They are part of the measurement.

A cereal aggregate can contain individual kernels with very different toxin concentrations. One particle may carry much more contamination than the next. If the laboratory takes a small portion from a coarse, poorly mixed aggregate, the portion can inherit the same hot-spot problem that affected the original lot, only at a smaller physical scale.

Grinding reduces the size of the units and helps distribute contamination throughout the material. It does not destroy the toxin or make a bad lot safe. It makes the analytical portion more likely to reflect the aggregate from which it was taken.

If the grind is too coarse, the extraction solvent may not interact uniformly with contaminated material. Large fragments can remain unevenly distributed, and a small test portion may contain too few of them. If the particle-size distribution is broad, the sample may segregate during handling: fine material settles differently from coarse material, and the final portion may no longer match the composition of the aggregate.

The consequence can be a low result, a high result, or simply poor agreement between replicate portions and laboratories.

Sieving is a control, not a formality

The 2023 USDA Federal Grain Inspection Service Mycotoxin Handbook specifies a particle-size requirement in which at least 60% of a ground sample passes through a U.S. Standard No. 20 sieve, corresponding to an 850 μm aperture. That is a minimum performance criterion for the relevant grain-testing procedures, not a universal specification for every commodity and method.

The laboratory should verify that its mill and sieve combination achieves the required distribution for the matrix being tested. A mill that performs well on maize may not produce the same result with a different grain, oilseed, nut, or processed feed. Wear, moisture, feed rate, and cleaning can also alter the final particle distribution.

The choice of equipment matters less than verification. Common laboratory mills may include rotor, cross-beater, or other systems designed for grain and feed preparation. The name of the mill does not prove that the sample meets the method requirement. The relevant evidence is the validated preparation procedure and the observed particle-size performance.

Why two competent laboratories can disagree

Inter-laboratory disagreement is often blamed on the chromatograph first. Sometimes the real difference appears earlier.

Two laboratories may use the same nominal HPLC method but differ in:

• how the aggregate was mixed;
• whether the sample was ground before subdivision;
• the proportion passing the specified sieve;
• the extraction ratio and mixing time;
• how difficult matrices were handled;
• the portion size submitted for analysis;
• whether modified forms were included in the target list.

This is why dispute testing should preserve the original material whenever possible and document each reduction step. Splitting a coarse or poorly mixed sample into two portions does not create two equivalent test materials. The subsamples can differ simply because contaminated particles were distributed unevenly.

A defensible preparation sequence generally keeps the aggregate intact until it has been mixed and reduced with suitable equipment. The laboratory should also be able to explain how it prevents segregation during transfer, grinding, storage, and final weighing.

Designing a Sampling Plan That Can Survive a Dispute

A sampling plan is not a generic instruction to collect several samples. It is a description of how a particular lot will be observed.

Before collection begins, define the lot using the information available at the point of control:

• commodity and product form;
• origin, supplier, and relevant production or storage identifiers;
• quantity and packaging or bulk presentation;
• loading and unloading arrangement;
• whether the material is continuous, static, bagged, or mixed;
• conditions that could create separate populations within the shipment.

If the material comes from separate origins, storage units, drying runs, or production dates, combining it into one lot may conceal meaningful differences. The lot boundary should follow the way the material was produced and handled, not just the number printed on a transport document.

Then define how increments will be distributed. A written plan should specify the route through the lot, the time or distance interval for moving material, the positions and depths for static material, and the equipment required. It should also specify what happens when the planned access is impossible. Replacing a systematic plan with a few convenient scoops should be recorded as a deviation, not silently treated as equivalent sampling.

The aggregate needs its own controls. It should be large enough for the applicable method and preserved through mixing and reduction. If a portion is sent to more than one laboratory, the division should occur after suitable homogenization and with equipment designed to produce comparable portions.

The final analytical request should match the decision. A buyer checking a regulatory limit may need a defined official-control method. A feed manufacturer assessing animal exposure may need a broader toxin panel, additional modified forms, or a different interpretation of the result. Those are related questions, but they are not identical.

What an Actually Useful Sampling Plan Looks Like

A practical plan for mycotoxin control should include the following elements:

• Define the lot before sampling. Treat the lot as an identifiable quantity with sufficiently common characteristics. A truck, bin, or barge may be the lot, but only when the material within it can reasonably be treated as one unit. The container alone is not the definition.
• Use incremental sampling across the lot. Collect increments systematically through the relevant volume or flow period. Do not rely on one top-layer grab or one convenient discharge point.
• Apply the current commodity-specific requirements. For EU official control, consult Commission Implementing Regulation (EU) 2023/2782, applicable from 1 April 2024, rather than relying uncritically on older references to Regulation (EC) No 401/2006.
• Build and preserve the aggregate sample. Use the required aggregate mass and incremental count for the commodity and lot size. Do not discard material simply because the full aggregate is inconvenient to transport.
• Homogenize before taking the laboratory portion. Reduce the aggregate only with a suitable, validated procedure that limits segregation and preserves comparability between portions.
• Verify particle size. For grain procedures covered by the relevant FGIS guidance, confirm that at least 60% of the ground sample passes the U.S. Standard No. 20 sieve, or follow the particle-size requirement specified by the applicable method.
• Clarify the analytical scope. Establish whether the laboratory measures parent mycotoxins only, selected masked forms, a defined sum, or a digestion-based estimate of potential availability.
• Record the process, not just the number. The certificate or accompanying record should identify the lot, location, date, operator, equipment, increment count, aggregate weight, deviations, reduction method, and chain of custody.
• Plan for disagreement before it happens. Retain enough properly handled material for confirmation or dispute testing, and make sure the retained portion is comparable to the submitted portion.

None of this requires exotic technology. It requires treating sampling as the load-bearing stage of the test rather than as a chore performed before the laboratory begins its work.

The instrument matters. The extraction chemistry matters. The analyst matters. But the laboratory can only measure the material it receives, and the material it receives is useful only when the sampling plan gives it a credible connection to the lot.

That is the point behind representative sampling for food safety. A precise answer from an unrepresentative sample is still the wrong answer.