How to read a fruit certificate of analysis: what each line proves

A certificate of analysis fruit importers receive from a supplier proves one thing: that a named sample from one lot was measured by stated methods and signed by one person. It is not a factory certificate. Read the lot code, sampling date, method number, LOQ, measurement uncertainty, ISO/IEC 17025 scope and signatory line by line, and write the decision rule into the contract.

What does a COA prove that a factory certificate does not?

A COA proves that someone measured certain parameters on a sample from a specific lot and got certain numbers. A factory certificate proves that an audited management system existed at a site on the audit date. The two documents answer different questions, and accepting one in place of the other is how a lot nobody tested gets received.

Three documents get confused on supplier pages. The specification is the acceptance limit both parties signed. The COA is the actual result for one lot, issued by the plant or a laboratory. In a dispute the specification is the yardstick and the COA is the evidence. The facility certificate is the third document, and it sits outside that pair entirely.

HACCP, the GFSI schemes (BRCGS, IFS, FSSC 22000, SQF), Halal and Kosher certificates are issued to a production site. GLOBALG.A.P. is issued to a farm or growing area. None of them is a lot result. The Centre for the Promotion of Imports from developing countries (CBI) wrote on 30 January 2024 that GFSI certification is “nowadays the minimum entry criteria for most customers” in the EU, which makes it a gate to the conversation, not evidence about a carton.

A certificate also has a scope naming product groups and technologies. A plant graded AA under BRCGS for canned pineapple is not thereby certified for IQF banana. Right certificate, wrong scope, and an EU customer still refuses the goods. You can check a facility certificate yourself in the BRCGS Directory, the IFS Database or the FSSC 22000 public register; for HACCP or ISO 22000, check that the certification body is itself accredited.

One phrase to discard on sight: “meets European standards” is not the name of any certificate.

The COA works the other way round. Any parameter absent from a COA is a parameter nobody measured.

Which fields does a usable COA carry, and how does each one fail?

A usable COA carries twelve fields, and each is a separate claim with its own failure mode. The first check is always the lot code against the delivery, because a COA for a different lot is paper. The table below is the order a quality manager should read them in.

The forged lot code is caught by comparing the test date with the production date and asking for the original lab report carrying the lab’s own sample ID. The quietly missing parameter is caught only by reading the COA line by line against the contract’s mandatory list, because the lines that go missing are the expensive ones: Alicyclobacillus, Howard Mold Count and the multi-residue pesticide panel. The same field-by-field reading applies to the document set for frozen durian and the paperwork that travels with it.

Which test method should sit beside each line?

Every quantitative line on a fruit COA should name a recognised method, because two COAs reporting the same number by different methods are not comparable. The methods below are the ones a buyer should expect to see printed beside each parameter on a frozen, dried or purée lot.

Two parameters have no standard at all. There is no Bostwick calibration standard and IFU publishes no consistency method, so two labs disagreeing on purée consistency is normal; the contract must fix sample temperature, volume, trough levelling and who calibrates. Colour measured as L*a*b* compares different quantities unless colour space, illuminant, observer and cuvette are all specified. Brix reporting for frozen fruit is covered in more detail in the IQF pineapple specification on cut styles and Brix.

For the lot itself, the sampling reference is Codex CXS 233-1969 at AQL 6.5, Inspection Level I, moving to Level II in a dispute. For pesticide-residue sampling of raw material, Codex CAC/GL 33-1999 takes 10 primary samples from lots over 500 kg and a laboratory sample of at least 1 kg.

Patulin is the line not to pay for on most tropical lots. Patulin is an apple and pear risk; it does not belong on a default COA for frozen or dried tropical fruit, but check the buyer’s juice specification before dropping it from a juice or concentrate COA.

What do “not detected” and “absent” actually mean?

“Not detected” means below the method’s limit of detection on the sample tested; a usable COA prints the LOD and LOQ beside the line. “Absent” means not found in the mass of sample tested. Neither means zero. Without the LOQ printed beside it, “not detected” cannot be judged; if the LOQ sits above the contract limit, the result says nothing about compliance with that limit.

For presence/absence pathogen tests there is no measurement uncertainty, but the probability of detection depends on sample size. Absent in 25 g and absent in 375 g are different tests, and a COA line saying “absent” with no mass is incomplete.

Whether the Listeria line applies at all depends on how the lot is declared for use, not on the fruit. The same IQF bag labelled “ready for smoothies” is ready-to-eat; labelled “ingredient, must be heated” it is not. That declaration decides which line the COA has to carry.

Does “ISO 17025 accredited” cover the test on your COA?

Not necessarily. ISO/IEC 17025 does not accredit a laboratory as a whole; it accredits specific tests on specific sample matrices over specific ranges. A lab that “holds 17025” can be unaccredited for the test you need on the matrix you need, and a COA can carry the accreditation logo at the top of the page over a line that is outside scope.

Ask the lab for its scope of accreditation and find the exact line covering the analyte, the method, the matrix (frozen fruit, purée or juice, not “food in general”), and the range and LOQ. In Vietnam the national accreditation body is BoA, and its laboratory accreditation numbers carry the prefix VILAS. Treat a lab’s 17025 claim as proven when the number can be looked up in the public directory.

Out-of-scope results still have technical value, but they carry much less weight in a dispute and the importing country’s authorities may refuse them. Whether an EU buyer accepts a Vietnamese laboratory’s COA or requires an EU laboratory is a contract term, settled with the buyer, not a fixed rule.

Who pays the lab matters as much as its scope. A lab chosen and paid by the plant is useful for day-to-day control, but it should not be the basis for accepting a high-value lot; for that, use a buyer-appointed lab or a third-party lab named in the contract. The facility certificate is a separate check with its own timing: VinFruits Global supplies the partner facility’s FSSC 22000, HACCP and Halal certificates with the quotation, so the scope can be checked against the product before a lot’s COA is ever issued.

Why do two accredited labs give two numbers for the same lot?

Because every quantitative result is x ± U, and two labs measuring the same lot will land at different points inside that spread. For pesticide residues the EU has written down how to decide between a result and a limit. A purchase contract has no such rule unless the parties write one.

“Compliance with the MRL must be checked by assuming that the MRL is exceeded if the measured value exceeds the MRL by more than the expanded uncertainty.”

Under EU guidance SANTE/11312/2021, the analytical quality control document for pesticide residues in food and feed, a lot is non-compliant when the measured value minus the expanded uncertainty exceeds the maximum residue limit, written x − U > MRL. The guidance sets a default expanded uncertainty of 50%, with coverage factor k = 2 at 95% confidence, for enforcement decisions. Its own worked example: an MRL of 1 mg/kg, a measured result of 2.2 mg/kg, U of 50%, gives 2.2 − 1.1 = 1.1, which is above 1.0, so the lot fails. The same arithmetic gives a different answer at 1.4 mg/kg against the same 1.0 mg/kg MRL: 1.4 − 0.7 = 0.7, which is below 1.0, so EU enforcement does not treat it as an exceedance. With a 50% default uncertainty, enforcement proof begins above twice the MRL. The 50% figure belongs to pesticide residues only and is not a general uncertainty for other COA lines.

The reverse risk runs against the buyer. A lot measured just under the MRL on the supplier’s COA can measure over it at the importer’s lab, so an internal specification should sit inside the MRL, not at it. EU retailers often set private residue limits tighter than the law, and those private limits are contract terms that get no uncertainty buffer at all.

A contract with no decision rule leaves a borderline lot undecided. “Exceeded when x > limit” fails it, which favours the buyer. “Exceeded when x − U > limit” passes it, which favours the seller. With nothing written, the dispute usually ends in a split loss or a discount. Inter-lab spread differs by parameter: pesticide residues carry the widest, quantitative microbiology commonly differs by around a log, which is why n/c sampling plans exist, and ICP-MS heavy metals sit lower than pesticides. For lead or cadmium, use the U the lab reports, not the pesticide default.

When two results conflict, work in this order:

1. Check the sample before the number: same lot, sampling point, storage, frozen or thawed. Most result disagreements are sample disagreements.
2. Check method number, matrix and basis, wet or dry.
3. Check that both tests are in 17025 scope with the same LOQ.
4. Apply U. If the two x ± U intervals overlap, the results do not contradict each other.
5. Compare with the limit using the contract’s rule.
6. If they still conflict, send the sealed retained sample to the referee lab named in the contract, for a final result.
7. The party whose result was wrong pays the referee cost.
8. Record it in the supplier file; three deviations in the same direction is a system, not chance.

Official controls already work this way. Regulation (EU) 2017/625, Article 35, gives an operator subject to official sampling the right to a second expert opinion at its own expense, including review of sampling and analysis records and re-analysis of the remaining sample, except where re-analysis is impossible or irrelevant, as with some microbiological tests. A commercial contract should copy that mechanism, with one exception kept explicit: for a presence/absence pathogen, a negative re-test on another portion does not cancel an original positive.

When does a COA stop being a private document?

When the destination country makes a laboratory result a condition of entry. For Vietnamese durian and jackfruit going to China, a COA counts only if it comes from a laboratory recognised by China’s General Administration of Customs (GACC), and a lot without one, or over the limit, is returned or destroyed.

Vietnam’s Plant Protection Department reports that 100% testing of durian and jackfruit consignments for cadmium and auramine O applies from 1 December 2024; another source dates it from January 2025. The limits are cadmium at or below 0.05 mg/kg under GB 2762, and zero tolerance for auramine O, an industrial dye once used to colour the husk. Sampling takes 2 fruit at random from 5 to 6 cartons per container, with registration at least 7 days ahead. Reported turnaround at the listed labs runs 1 to 2 days off-peak and 4 to 8 days at peak.

The list of recognised labs is the document that matters, and it moves. Tiền Phong reported on 6 May 2026 that only around 15 units nationwide could test cadmium, and the bottleneck held up the jackfruit trade. Published counts from June to August 2026 disagree with each other, so no count is given here as settled. The authoritative version is the dated list published by the Plant Protection Department at ppd.gov.vn, last updated 14 August 2026. A COA for this purpose is valid only from a lab on that list on the date of testing. The pressure behind the regime is real: Thương hiệu & Công luận reported in May 2026 that around 30% of durian and jackfruit samples from the Mekong Delta exceeded the cadmium threshold. Thailand, by comparison, has more than 300 on-farm testing points recognised by Chinese customs and is checked at 30% rather than 100%.

The EU works differently. Under Regulation (EU) 2019/1793, goods in Annex I face increased-frequency border checks, and goods in Annex II must travel with an official health certificate and analytical results from a recognised laboratory for every consignment, produced before export. Separately, Vietnamese fresh lines in the consolidated version as of 27 March 2026 were listed at border-check frequencies of dragon fruit 30%, chilli 50%, okra 50% and durian 20%; confirm in the consolidated text which annex a given line sits in before deciding what must travel with it. The same durian COA is therefore a contractual document in Rotterdam and an entry condition at the Chinese border.

What should the purchase order say?

The purchase order should list the parameters the COA must carry, the method and reporting format for each, who tests, how the retained sample is kept, and the decision rule. The clauses below can be pasted into a contract annex and filled in.

1. Mandatory COA parameters. “Each lot shall be accompanied by a COA reporting, parameter by parameter, the following tests: [list]. Any parameter not reported is treated as not tested and the lot as non-conforming.”
2. Reporting format. “Each result shall state the method number, LOQ, expanded measurement uncertainty, unit and basis (frozen, thawed or dry matter). Presence/absence results shall state sample mass. ‘Pass’, ‘complies’ and ‘not detected’ without LOQ are not accepted.”
3. Lot identity. “The COA shall carry the lot code of the delivered goods, production date, sampling date and test date, and the laboratory’s internal sample ID. Testing shall be no more than [60] days before delivery.”
4. Accreditation. “Each test shall fall within the laboratory’s ISO/IEC 17025 scope for the stated matrix, with the accreditation number printed and out-of-scope tests marked.”
5. Retained sample. “At delivery, a retained sample shall be drawn, sealed and signed by both parties, stored under label conditions with a continuous temperature record for frozen goods, held by [party] until shelf life plus 6 months.”
6. Referee laboratory. “Disputes shall be referred to [named laboratory, address], or failing it [named back-up], testing the sealed retained sample by the identical method.”
7. Decision rule. “A result exceeds a limit when [x > limit] / [x − U > limit]. For presence/absence pathogens, an original positive stands regardless of later negatives.”
8. Deadlines and cost. “A referee test must be invoked within [n] days of receipt of results. The cost is borne by the party against whom the referee finds. The referee result is final and binding.”

Field 7 is the one that decides who wins. Write it before the lot ships.

Common questions

Is an FSSC 22000 certificate enough without a COA?

No. FSSC 22000, BRCGS, IFS, HACCP and Halal certificates are issued to a production site after an audit of its management system, for the products named in the scope. None of them reports a test on the lot you are receiving. A COA reports measured results for one lot. You need both, and the certificate’s scope has to cover the product line you are buying.

Does not detected on a COA mean zero?

No. Not detected means the result was below the method’s limit of detection on the sample tested; a usable COA prints the LOD and LOQ beside the line. Without the LOQ printed beside it the line cannot be judged, and if the LOQ sits above your contract limit the result says nothing about compliance. For pathogens, absent also needs the sample mass, because absent in 25 g and absent in 375 g are different tests.

How old can a fruit COA be?

No single legal age applies to every contract. The USDA Commodity Specification for Frozen Fruit of June 2017 requires inspection no more than 60 days before delivery, which is a usable reference. More important is that the sampling date and test date fit the production date of the lot code on the delivery. A COA dated before the lot was produced describes another lot.

Can a supplier’s own lab issue the COA?

It can, and plant QC results are useful for daily control. For accepting a high-value lot, buyers rely on a buyer-appointed or contract-named third-party laboratory with the test inside its ISO/IEC 17025 scope. Check the lab name, accreditation number and signatory, because a plant’s in-house sheet formatted like a lab report is a common substitution.

What happens when my lab and the supplier’s lab disagree?

First check whether both tested the same lot, sampling point and state, frozen or thawed, by the same method. Then apply each result’s measurement uncertainty: if the two x ± U intervals overlap, the results agree. If they truly conflict, the referee lab named in the contract tests the sealed retained sample, applies the contract’s decision rule, and the party found against pays.