Testing does not make food safe. Control makes food safe, and testing tells you whether the control worked. Every well-run programme starts from that order, because the alternative, testing your way to a decision on a lot you have no other information about, is statistically weak and financially ruinous at the same time.
This guide covers four questions in sequence: what to test, how often, where to have it done, and how to read what comes back.
What to test: work down from the hazard analysis
The parameters worth paying for are the ones your hazard analysis identifies as significant and not fully controlled by a preceding step. Everything else is either a specification parameter you check because the contract says so, or habit.
Four families cover almost all of it.
Microbiological. For frozen fruit the dominant concerns are the ones a freezing step does not remove, because freezing is not a kill step. Union food safety criteria and process hygiene criteria for particular categories sit in the microbiological criteria regulation, and the categories to which they apply are defined there rather than by analogy. Salmonella testing and PCR-based screening are the usual routes, with cultural confirmation of presumptive positives.
Chemical contaminants. Maximum levels sit in Commission Regulation (EU) 2023/915. Which contaminants matter depends entirely on the commodity: mycotoxins in nuts, dried fruit and grains; heavy metals in leafy and root material and in some mushroom species; process contaminants where a heat step is involved. ICP-MS covers the metals; LC-MS/MS covers most mycotoxins.
Pesticide residues. Regulation (EC) No 396/2005 sets maximum residue levels, and a multi-residue screen by GC-MS/MS and LC-MS/MS is the standard instrument. The screen is only as good as its scope: a panel of several hundred actives will not find a substance outside the panel, and the actives in use in the country of production are the ones the panel needs to cover.
Composition, authenticity and physical. Moisture, water activity, proximate composition, sugar profile, sensory and defect counts against a defect classification. These decide commercial acceptability rather than safety, and they are the parameters that generate most claims.
Write each chosen parameter into the product specification with a method reference and a limit. A parameter without a method is not testable and a limit without a method is not enforceable.
How often: frequency is a risk decision, not a ritual
Batch-by-batch testing of everything is the most expensive way to learn the least. The workable structure has three tiers.
- Approval testing. A full panel on the first production lots from a new supplier, a new origin or a new season, covering everything in the specification. This is the baseline against which later results are judged.
- Verification testing. A reduced panel at a defined frequency, one lot in N or a fixed number per season, chosen so that a drift is caught within an acceptable window. Frequency rises with the variability of the raw material and falls with the maturity of the supplier relationship.
- Triggered testing. A full or targeted panel when something changes: a new grower group, a border rejection anywhere in the category, an alert on the origin, a process change, a complaint, or a result trending toward a limit.
Two commodity-specific adjustments. Where the origin is subject to temporarily increased official controls at the Union border, the testing plan should mirror what the border will do rather than discovering it there. And for contaminants distributed heterogeneously, aflatoxin in nuts being the classic case, frequency matters far less than sampling design, for the reason set out next.
Sampling dominates the result
For a heterogeneously distributed contaminant, most of the total error in a result comes from the sampling step, not from the laboratory. A tonne of walnut kernel can carry its entire aflatoxin burden in a handful of kernels. Whether those kernels reach the analyst is decided by how many incremental samples were taken, from where, how they were combined, and how the aggregate sample was ground before subsampling.
That is why sampling is legislated rather than left to the laboratory. Commission Implementing Regulation (EU) 2023/2782 prescribes sampling and analysis for mycotoxin control, replacing Regulation (EC) No 401/2006. Commission Directive 2002/63/EC does the same job for pesticide residues. Both specify incremental sample counts by lot size, aggregate sample mass, sublot structure and comminution.
The practical consequence for a buyer is short: a certificate with an impressive limit of quantification and no statement of how the sample was drawn is not evidence about the lot. It is evidence about the sample. Require the sampling reference on the certificate, and where the value at risk justifies it, require that sampling be witnessed or performed by the laboratory or an inspection body rather than by the seller.
For statistical acceptance of visual and count-based attributes, defect inspection is normally run to a published sampling scheme indexed by acceptance quality limit, with a general inspection level chosen for the lot size and a switching rule that tightens after failures. Naming the scheme, the level and the acceptance quality limit in the contract removes an entire class of argument at intake.
Where: accreditation, and what it actually covers
ISO/IEC 17025 is the international standard for the competence of testing and calibration laboratories. The 2017 edition is process- and risk-based, and it combines management system requirements with technical requirements: personnel competence, method validation, equipment calibration, metrological traceability, measurement uncertainty and participation in proficiency testing.
The point most often missed is that accreditation is granted against a scope, not to a building. A laboratory is accredited for specified methods on specified matrices. A laboratory accredited for aflatoxins in maize is not thereby accredited for aflatoxins in dried figs, and a report from that laboratory on figs is an unaccredited result even though the letterhead carries an accreditation mark. Ask for the scope document, check that your method and your matrix appear in it, and check the expiry.
Cross-border acceptance runs through the ILAC Mutual Recognition Arrangement. Accreditation bodies that have been peer-evaluated against ISO/IEC 17011 sign the arrangement, and results from laboratories they accredit are recognised by the other signatories. That is what allows a Ukrainian laboratory’s report to be accepted by a German buyer without repeat testing, and it is worth confirming that the accreditation body itself is a signatory, not only that the laboratory holds a certificate.
Inspection bodies, which is what a pre-shipment inspection provider is, are accredited to ISO/IEC 17020 rather than to ISO/IEC 17025. The two are different arrangements and a body may hold one and not the other.
The choice between running the work in-house and buying it is treated separately in in-house against third-party laboratory testing.
Reading a certificate of analysis
A certificate of analysis should let a reader reconstruct what was done. Check these, in this order.
- Lot linkage. Does the sample identity tie to a production lot code you can find on the pallet, or only to a date? A certificate that cannot be tied to a lot is a marketing document.
- Sampling. Who drew the sample, when, against which sampling standard, and how many increments. If this section is absent, the certificate covers the sample only.
- Dates. Date of sampling, date of receipt, date of analysis. A long gap between drawing and receipt invalidates several parameters outright, peroxide value and microbiology among them.
- Method. A named, referenced method, not “in-house method” alone. Where an in-house method is used, its validation status should be stated.
- Limit of detection and limit of quantification. A “not detected” is meaningless without the limit it was not detected above. Contractual limits below the laboratory’s limit of quantification cannot be demonstrated.
- Measurement uncertainty. ISO/IEC 17025 requires laboratories to evaluate it. It is the range within which the true value is expected to lie, and it is why a single result marginally above a specification limit is not automatically a failure. How uncertainty is applied to a compliance decision differs between regulatory enforcement and contractual acceptance, which is precisely why the contract should say which convention applies.
- Accreditation marks. Present on the report, and applied to the specific results within scope. Many reports mix accredited and non-accredited parameters and flag the difference in a footnote.
- Signature and authorisation. An authorised signatory, not an unsigned system export.
Putting it in the contract
The testing plan is worth nothing if it is not written into the commercial agreement. Six clauses cover most disputes.
- The parameters, methods and limits, by reference to the specification rather than repeated.
- The sampling standard and who draws the sample.
- Which laboratory, or which class of laboratory, and whose accreditation scope must cover the method and matrix.
- Who pays for routine testing and who pays when a result fails.
- Retained samples: how many, held by whom, for how long, under what conditions.
- An umpire clause naming a third laboratory whose result binds both parties in a dispute, with the cost following the outcome.
Vorezan does not operate a laboratory, does not issue certificates of analysis and does not accredit laboratories. Testing programmes and the interpretation of results should be established with the operator’s own technical function and with a laboratory whose accreditation scope covers the specific method and matrix.
