NIR (near-infrared) spectroscopy sends light into a sample and measures the light that comes back, giving a fast, non-destructive fingerprint of the product's composition. That fingerprint does not count as a result on its own: to mean anything it has to be compared through a properly built calibration model and against reliable reference samples. That is why NIR is most useful at the screening stage; for a disputed batch, the final decision is left to laboratory analysis.
How does NIR spectroscopy work?
Light in the near-infrared region is absorbed by particular bonds in the molecules of a food. These are mainly carbon-hydrogen, oxygen-hydrogen and nitrogen-hydrogen bonds; in other words the bonds carried by major components such as fat, water, carbohydrate and protein. Which wavelengths are absorbed, and how strongly, makes up the product's spectrum.
That spectrum is not read directly. The bands overlap, and the share of a single component cannot be made out by eye. What turns the spectrum into a number, or into a "matches / does not match the expected profile" decision, is multivariate statistical modelling. These models are built from a large number of reference samples of known composition. Every NIR result is in effect a statement of how closely the measured sample resembles those references.
What is NIR used for in food analysis?
The method is fast, does not damage the sample and needs no chemical consumables. That is why it has long been used in the food industry for:
- Routine measurement of major components such as moisture, fat, protein and dry matter.
- Quick checks at raw-material intake of whether a batch matches the expected profile.
- Comparing batches from the same supplier against one another over time.
- Sorting out batches suspected of adulteration before they are sent to the laboratory.
The last item needs the most care. NIR can show that a batch is different; it cannot always say why it is different or by how much.
Can portable NIR devices replace the laboratory?
Handheld and portable NIR devices have become widespread in recent years. The World Health Organization, too, lists handheld spectrometers among the tools used against substandard and falsified medical products.
Peer-reviewed literature is clear about their limits. A 2021 review of applications in milk, cheese and dairy powders reports that handheld devices are not as sensitive as benchtop NIR instruments at detecting adulterants in milk powder. According to the same review, spectra from miniaturised devices can be noisier, the models built on them need ongoing maintenance, and it is often hard to know how reliable a prediction is. The authors position the portable device as a tool that complements the benchtop instrument.
On that basis, portable NIR suits giving a fast answer at intake to the question "should this batch be looked at more closely?" Whether the batch is adulterated, and at what level, is established by laboratory analysis.
What does the reliability of an NIR result depend on?
The same device can give very different results in different set-ups. What decides it:
- The reference sample set. A model recognises only the variety covered by the samples it was built on. We explain how that set is built and grown in the reference library post.
- The accuracy of the reference values. Errors in the laboratory values used to build the model carry straight into the prediction.
- Re-validating the model. When a new harvest, a new supplier or a new device arrives, the model is not simply kept in use; it is first tested against new samples.
- Measurement conditions. Temperature, sample preparation, and whether measurement is through packaging or on an open sample all change the result.
- The size of the difference sought. Catching a low-level addition is much harder than catching a crude substitution.
The first three items concern the data behind the device, not the device itself. That is also the most common reason two users get different results from the same instrument.
When does detecting adulteration with NIR get harder?
When the added substance is chemically similar to the main component, the difference in the spectrum stays small. That is the case when one vegetable oil is blended into another, or one sugar source into a sugar syrup. With substitutions of this kind, screening can only catch high proportions.
Products with high natural variation are the second hard group. In honey, olive oil and spices, region, harvest period and processing method shift the spectrum by themselves. If the reference set does not cover that variety, the model may flag a genuine batch as suspicious, or mistake a real deviation for natural difference.
So a "does not match" result is used to send the batch to the laboratory. A "matches" result does not show that a low-level, chemically similar addition is absent.
When is laboratory analysis needed?
A screening result is not enough when:
- The result will lead to rejecting a supplier or enforcing a contract.
- A report is to be submitted to a buyer, an auditor or an authority.
- The screen shows a difference but its source is unknown.
- The product carries an origin or geographical indication claim.
In these cases the screen's role is to choose which batch to sample. Whether the result stands up to a challenge depends on taking the sample properly and setting aside a witness sample; see what makes a sample valid.
How does Molevex divide screening and the laboratory?
Molevex measures the product's own composition without adding anything to it, and compares it with a sample you know to be correct. The quick look in the field decides which sample goes on for further examination; the result you present to your buyer and auditor rests on laboratory confirmation. See the technology page for the details of the approach.


