Indiquez vos coordonnées pour télécharger le document.
Le fichier s’ouvre dans Google Drive : vous pouvez le télécharger ou l’imprimer.
Which batch, how many units, how often, which criterion: the method to write a reproducible, defensible sampling plan for your raw materials, finished products and suppliers.
You have to decide what you have sampled and analysed on your raw materials, finished products and at your suppliers: the sampling plan is the document that sets these rules. It answers three concrete questions - on which batch, how many units, and from what result the batch is accepted or rejected - and it remains your responsibility, including when sampling is entrusted to a service provider. This guide is aimed at food manufacturers who build and justify this plan, not at the laboratory that carries out sampling and testing.
The sampling plan is the document that turns an intention to check into reproducible actions: for a category of batches, it specifies what is sampled, in what quantity, at what time and by which rule the result leads to acceptance or rejection. The control plan covers the whole monitoring system and contains the sampling plan as one of its operational chapters.
For a family of materials or products, the sampling plan usually fits on one page and covers six elements: the sampling unit, the number of units to take, how to select them, the frequency, the acceptance criterion and the person responsible for sampling as well as for the decision. It must be understandable by an operator who did not take part in writing it, and applicable without interpretation.
It is neither a test protocol nor an analytical method: the sampling plan does not say how to measure a substance, it says what to sample and what is done with the result.
The control plan lists all the points monitored - receipt of materials, work in progress, finished product, environment - with the parameters, methods, limits, frequencies and handling of deviations. The sampling plan is the executable part: it turns each control point into sampling actions.
A control plan that states “microbiological analysis on receipt” without a sampling plan cannot be applied: nobody knows how many units to take, from which batch, or what to do with the result. For building the document as a whole, see our page Quality control plan: method and examples in the food industry.
Six parameters are enough to make a quality sampling plan usable. If one is missing, actual sampling fills the gap, and rarely in the same way twice.
| Parameter | Question to settle | What the plan must state | Point of attention |
|---|---|---|---|
| Batch size | Which homogeneous set do I consider a batch? | An explicit, identifiable definition: supplier batch number, delivery, production day or pallet, depending on the convention chosen | A batch reconstructed after the event, when the non-conformity appears |
| Sample size | How many units should be taken from this batch? | A number (example: 5 units per batch), with the associated selection rule | Always sampling the most accessible units |
| Frequency | Every how many batches or receipts? | An interval and its trigger (at every receipt, per campaign, after a change of supplier), for example at every receipt for a supplier on probation | Frequency confused with sample size |
| Acceptance criterion | From what result is the batch outside the criterion? | The applicable limit and its written source: regulatory requirement, customer or supplier specification, justified internal threshold | An internal threshold that refers to no reference document |
| Owner | Who samples, who decides, who records? | A function, not a name: sampling by the goods-in operator, decision by the quality manager (example) | An implicit responsibility, and therefore missing on the day someone stands in |
| Sampling traceability | How do you link the sample to the batch, the report and the product? | Sample ID, date and time, operator, location, storage and transport conditions, link to the report | A sample ID not carried over to the test report |
The same table works upstream: a raw material control plan is read line by line, one line per purchased item, starting from the supplier batch. The figures in each line are not copied from one site to another: they are decided, then justified in writing, as the rest of this article explains.
From risk to the batch decision
Order matters: risk first, then the batch, then the decision rule before the number of units.
Construction follows a precise order: risk first, then the batch, then the decision rule before choosing the number of units. This order avoids plans that look rigorous but detect nothing useful.
For each material, product or supplier, identify the characteristic or hazard you are trying to detect, and the consequence of an undetected defect. A critical item, a supplier already in difficulty or a sensitive process does not deserve the same plan as a stable material from a supplier assessed for years. This prioritisation must exist in writing: risk analysis, history of non-conformities, supplier assessment results.
A batch must be homogeneous, identifiable and traceable. If you cannot name the batch, the analysis result cannot be linked to any quantity of product: it becomes information with no operational consequence. The batch reference must appear both on the sampling record and on the sample itself.
The sampling unit (a sack, a carton, a sales unit) must be smaller than the batch and large enough to allow the planned test. The selection must then be described: random sampling, stratified sampling across several areas of the batch, or sampling of units chosen for their representativeness. An instruction such as “take a few sacks” is not a rule.
The acceptance criterion can take three forms: zero non-conforming units for a critical hazard, a tolerated number of non-conforming units for a minor defect, or a numerical threshold for a measurement. For microbiology, Annex I to Regulation (EC) No 2073/2005 sets for certain foods both the plan (n, c) and the limits (m, M): this is the reference to use where it applies. In all cases, write down the source of the criterion and the action planned when the criterion is not met: goods-in block, supplier complaint, re-analysis, documented concession.
This question is dealt with in the next section: it determines the type of plan, and therefore how to decide, not just how to count.
Who samples, with what clean equipment, in what container, with what identification, within what time and at what transport temperature, for which recipient. This is where quality control sampling succeeds or fails: a sample that is correctly identified, stored and sent is worth more than a sophisticated plan applied approximately.
Run the plan over a limited period, then look at what it produces: non-conformities detected, non-conformities missed, time spent, cost of analyses, supplier disputes. These observations lead to adjusting the frequency or number of units, and to dating the revision. A plan never revised ends up matching neither actual purchases nor your customers' requirements.
The choice depends on the nature of the judgement you have to make, not on the team's preference. An attributes plan decides on a unit-by-unit finding of conformity; a variables plan decides on a numerical value and its variability.
| Question | By attributes | By variables |
|---|---|---|
| What is assessed | Each unit is conforming or non-conforming, or defects are counted | A continuous quantity, compared with a limit |
| Data required | A simple finding, reproducible by several operators | A numerical result, with known variability |
| Family of plans | ISO 2859, sampling procedures for inspection by attributes | ISO 3951, sampling procedures for inspection by variables |
| Conditions of use | Defect observable without measurement | Process under statistical control, distribution close to normal |
| Examples in the food industry | Packaging integrity, appearance, presence of foreign bodies | Content, moisture, pH, result of a test report |
| Main limitation | Little information on the margin relative to the limit | Requires usable data and a well-understood method |
In both families, standard tables are chosen from an AQL (acceptance quality limit), that is the quality level still considered acceptable for a given plan. The AQL is used to select a plan, not to qualify an isolated batch: it is a design parameter, to be decided and written down in your document just like the sample size.
In practice, the two approaches often coexist: inspection by attributes at goods-in, then use of the numerical result in the test report for the decision on batch compliance. What matters is that each decision refers to a written rule and an identifiable criterion source. The goods-in approach is detailed on our raw material incoming inspection page.
Three mistakes recur in almost every plan reworked after an audit or a non-conformity. They are not about statistics, but about how the plan is written and applied.
Always sampling the top pallet, the carton already opened or the most accessible area produces a result that describes the place sampled, not the batch. The defect is corrected in the wording of the plan: describe the selection rule and check on the ground that it is followed, including at the end of the day and during peaks of activity.
Without a sample ID, a date and a batch reference, the result cannot be linked either to the product delivered or to the supplier concerned. The traceability required by Article 18 of Regulation (EC) No 178/2002 does not stop at commercial documents: it is also checked along the path between the sample and the test report.
Writing “one check per delivery” does not say how many units to take; writing “three units” does not say how often. These two parameters answer two different questions - how many batches to cover, and how much confidence to place in each batch - and must appear on two separate lines of the plan.
A plan can be perfectly written and no longer match anything: a purchased item that has changed supplier, a modified pack format, a new regulatory parameter. A forgotten revision date is one of the most frequent audit findings, because it is immediately visible on the document.
An auditor does not read your plan for pleasure: they start from a result, trace it back to the sample and then the batch, and check that the decision taken matches a written rule. The elements to keep follow from this path.
Two texts frame this system without replacing it. Regulation (EU) 2017/625 deals with sampling carried out as part of official controls (Articles 14, 34 and 35): it does not exempt operators from defining their own plan. And when sampling is carried out by the laboratory, ISO/IEC 17025:2017 (§ 7.3) requires it to have a sampling plan and method available at the sampling site: two documents to request, then compare with your own.
Tracklab comes in after sampling: the platform gathers test reports from your suppliers and laboratories, extracts the results and calculates for each parameter a compliance status against your criteria and the four European regulations it builds in and keeps up to date (pesticides, contaminants, microbiological criteria, plastic materials in contact with food). Each calculation keeps a trace of the threshold, regulation, its version and its date: the history is traceable and auditable. In the event of a deviation, a re-analysis or a request to the supplier can be recorded on the same report.
Tracklab does not carry out sampling and does not decide on batch acceptance: that decision stays with your quality teams. It collects reports, calculates the status of each parameter against your criteria, keeps “not detected” results and tracks the coverage of your analyses per product × supplier pair. For details, see Tracklab's LIMS software page.
If you want to see how your acceptance criteria apply to your own reports, you can book a meeting with the Tracklab team.
The texts below underpin the requirements cited on this page; always consult their current consolidated version on EUR-Lex.
A sampling plan is judged on its reproducibility: two operators must sample and decide in the same way.
Improvised sampling: Always the top pallet or the opened carton
Written plan: A selection rule described and checked on the ground
Improvised sampling: A sample with no ID or batch reference
Written plan: A traced link between sample, batch, report and decision
Improvised sampling: An internal threshold that refers to no document
Written plan: A dated acceptance criterion, with its source
Not necessarily one page per item: the plan can be organised by family of risks and purchasing conditions. However, each purchased item must be explicitly linked to a line of the plan, with its reference batch and criterion. An orphan item is a control point that exists only on paper.
There is no universal value: the number depends on the risk, the homogeneity of the batch, the precision of the test and the cost of the analysis. ISO 2859 and ISO 3951 provide usable tables, provided the AQL and inspection level are set first; for certain microbiological criteria, Regulation (EC) No 2073/2005 sets n and c directly. Without a standard table, the manufacturer writes its rule and justifies it; until it is decided, the plan line remains incomplete and cannot be applied.
That is a plan decision, not a given. Using the supplier's report as the basis for decision assumes an assessed supplier, known methods and a defined supply frequency; otherwise, verification analyses must be planned. In both cases, the frequency and acceptance criterion must be written down.
Yes, when it carries out the sampling itself: ISO/IEC 17025:2017 (§ 7.3) requires it, with a method available at the sampling site. If sampling is done by your teams, your plan applies, and the laboratory only deals with the sample received.
The interval is set by the manufacturer, according to how stable its purchases and processes are. Triggers are more useful than an annual date: repeated non-conformity, change of supplier or formulation, process change, new customer or regulatory requirement. Each review must be dated and traced.
Tracklab gathers reports from your suppliers and laboratories and calculates the compliance status of each parameter, against your criteria and the regulations. The decision on the batch stays with your quality teams.
0.02 mg/kgCompliant0.09 mg/kgAlert0.015 mg/kgNon-compliantAbsent /25 gTo be determinedIllustration, fictitious data.
Their quality and purchasing teams collect, control and leverage their supplier data with Tracklab.
companies supported
suppliers referenced
products and ingredients
“Tracklab has eliminated the blind spots that existed in our document monitoring. Every document is analysed, every change is detected and every decision is traceable. This has enabled us to take a real step forward in terms of professionalism and credibility, particularly with auditors and customers.”

Independent assessment of our cybersecurity framework and of the protection of the data entrusted to Tracklab.
They use Tracklab







Platform
Resources