How to Measure Smell: From Human Panels to Instruments

Measuring smell relies on two fundamentally different strategies: trained human noses and chemical instruments, each capturing something the other misses. Human sensory panels remain the gold standard for judging whether something actually smells bad (or good, or different), because no instrument truly replicates the subjective experience of sniffing. But human noses are inconsistent, and they get tired. Instruments like electronic noses and gas chromatography systems bring repeatability and chemical specificity, yet they struggle with the perceptual side of odor. In practice, most serious odor measurement programs combine both approaches, and the field is still working out how to bridge the gap between what a machine detects and what a person perceives.

Human Sensory Panels and Dynamic Olfactometry

The most established method for measuring odor strength uses groups of screened human sniffers. In Europe, the standard approach is dynamic olfactometry, formalized in the EN 13725 standard, which has become the preferred method for evaluating odor emissions from industrial and agricultural sources.¹ The technique works by progressively diluting an odorous air sample with clean air and presenting it to panelists through a sniffing port. Each panelist indicates when they can first detect an odor. The number of dilutions needed to reach that detection point gives you the odor concentration, expressed in “odor units.”

Panelists are not random volunteers. They are screened for their sensitivity to a reference compound, n-butanol, and must fall within a defined range of detection thresholds. The idea is that if everyone on the panel has a similar baseline sensitivity to this one chemical, their judgments about other odors will be reasonably comparable. The EN 13725 standard links its quality criteria for precision and accuracy to n-butanol, assuming that compliance with those criteria transfers to environmental odors more broadly.¹

This assumption, however, is shakier than many practitioners realize. A study examining datasets from three odor laboratories found that human sensitivity toward n-butanol did not correlate well with sensitivity toward other odorants or complex odor mixtures. Even structurally related compounds showed poor correlation. The researchers concluded that harmonizing panelists based on n-butanol does not result in harmonization for other odors or environmental samples.² This is a genuine weak spot in the system: a panel might pass all its quality checks for n-butanol while still producing inconsistent results for the pig-farm odor or wastewater plume it was actually assembled to evaluate.

Why Human Panels Are Still Hard to Replace

For all their limitations, human panels capture something instruments cannot: the actual experience of smelling. A gas chromatograph can tell you exactly which volatile compounds are present and in what concentrations, but it cannot tell you whether the mixture smells like rotten eggs or fresh bread. Human perception integrates hundreds of simultaneous chemical signals into a single qualitative impression, and that impression is what matters when communities complain about a landfill or a regulator decides whether an emission standard is being met.

The challenge is that human olfactory sensitivity varies enormously. Genetics, age, sex, medical history, infections, head trauma, neurodegenerative conditions, and emotional disorders all contribute to differences in how well people smell.³ Even recent odor exposure changes sensitivity: research in mice has shown that the olfactory bulb’s output is shaped by odors encountered in the previous thirty seconds or so, with adaptation effects that are concentration-dependent and not simply a matter of breathing patterns.⁴ In practical terms, a panelist who has been sniffing samples all morning does not respond to the afternoon samples the way they did at eight a.m.

And then there is culture. The assumption that odor naming and perception are universal turns out to be wrong. While English speakers have notoriously few dedicated smell words and find naming odors difficult, many other languages have sizable smell vocabularies, and in some cultures odor talk is frequent and odor naming comes much more easily. These different ways of talking about smells appear to shape aspects of olfactory cognition itself. When you ask a sensory panel to describe an odor character, the language and cultural framework of the panelists can influence what they report, adding a layer of variability that no screening protocol currently accounts for.

What Panels Actually Measure

Odor has multiple dimensions, and panels can be asked to evaluate several of them:

Detection threshold: the lowest concentration at which an odor can just barely be detected, but not necessarily identified. This is what dynamic olfactometry primarily measures.
Recognition threshold: the concentration at which a person can identify what the odor smells like, not just that something is there.
Intensity: how strong the odor seems at concentrations above the threshold.
Character: qualitative descriptions of what it smells like (musty, fruity, sulfurous, etc.).
Hedonic tone: how pleasant or unpleasant the odor is perceived to be.

The relationship between odor concentration and perceived intensity is not linear. Several mathematical models have been tested. The widely used Weber-Fechner law, which assumes intensity rises with the logarithm of concentration, does not always fit real data well. Researchers evaluating swine-building odors found that while the Weber-Fechner law was a poor fit, both a power-law model and the Beidler model described the data effectively, with the Beidler model performing best.⁵ The practical takeaway is that doubling the chemical concentration of an odorant does not double its perceived smell strength, and the exact relationship depends on the substance.

Odor mixtures add another layer of complexity. When two odorants are combined, the result is not simply the sum of their individual smells. Depending on the concentrations involved, one component can suppress another (masking) or the combination can smell stronger than either would alone (synergy). Electrophysiological studies have confirmed that these are not just subjective impressions: the brain’s early neural responses differ measurably between synergy and masking conditions.⁶ This means that analyzing a mixture’s individual chemicals and trying to predict its overall smell from those components is unreliable without accounting for perceptual interactions.

Field Olfactometry

Laboratory panels work well when you can collect a sample in a bag and bring it indoors, but sometimes you need to measure odor right where people are experiencing it. Field olfactometers solve this by mixing ambient air with carbon-filtered clean air in controlled ratios on the spot, letting a sniffer determine how many dilutions are needed before the odor disappears. The metric is the dilution-to-threshold ratio.

The first field olfactometer, called a scentometer, was developed in the late 1950s with funding from the U.S. Public Health Service. A more modern device, the Nasal Ranger, was introduced in 2002 and has been shown to provide consistent measurements of livestock-farm odor comparable to those from laboratory dynamic olfactometry.⁷ Field olfactometry is economically attractive and gives immediate readings, but it has real drawbacks. Sniffers working outdoors are exposed to the ambient environment before testing begins, leading to odor fatigue. Temperature, humidity, and wind speed change throughout the day. And there is some concern that sniffers who can see the emission source (a feedlot, a composting facility) may not remain fully objective about what they are smelling.⁷

Gas Chromatography Paired with Human Sniffing

One of the more elegant approaches to odor measurement marries an instrument with a human nose at the same time. Gas chromatography-olfactometry-mass spectrometry (GC-O-MS) separates a complex mixture into individual chemical compounds using a gas chromatograph, then splits the output so it flows simultaneously to a mass spectrometer (which identifies the chemical) and a sniffing port (where a trained person evaluates the odor). This lets researchers match each chemical peak to a smell impression, pinpointing exactly which compounds in a food or environmental sample are responsible for what you actually perceive.⁸

GC-O-MS has become a workhorse in food science, where it is used to map aroma-active compounds, identify key flavor contributors, and investigate how processing or storage changes a product’s smell profile. A study on cold-brew coffee, for example, used GC-O-MS combined with two different extraction methods to characterize the volatile compounds responsible for its distinctive aroma.⁹ The human sniffer at the port adds something that the mass spectrometer alone cannot provide: whether a particular chemical, at the concentration present, actually contributes to the smell experience. Many volatile compounds that show up as large peaks on a chromatogram have no perceptible odor at their natural concentrations, while trace compounds with tiny peaks can dominate the smell.

Electronic Noses

Electronic noses attempt to mimic human olfaction without a human in the loop. The most common type uses an array of metal oxide semiconductor (MOS) sensors, each of which responds to a broad range of volatile organic compounds but with different selectivity. When exposed to an odor, the array generates a pattern of electrical resistance changes across its sensors, creating what amounts to a fingerprint for that odor. Pattern-recognition algorithms then classify the fingerprint against a database of known samples.¹⁰

The sensors do not behave independently. Each one responds most strongly to certain chemical groups but also shows overlapping sensitivity to others, and it is this overlap that gives the array its discriminating power.¹¹ The concept is loosely analogous to how human olfactory receptors work: no single receptor type uniquely identifies a molecule, but the combinatorial pattern across many receptors does. The difference is that an e-nose typically has a few dozen sensors at most, whereas the human nose deploys several hundred receptor types.

The strength of electronic noses is consistency and speed. They do not get tired, they do not have good days and bad days, and they can run continuously on a production line. Their weakness is that they require training against known reference samples for every application. An e-nose trained to distinguish spoiled peanuts from fresh ones cannot spontaneously tell you what a new, unfamiliar odor smells like. And environmental factors like humidity and temperature can drift sensor baselines, requiring careful data preprocessing.

How Human Noses and E-Noses Compare

A direct comparison of a human sensory panel and an electronic nose evaluating wine volatiles found that the human nose was more sensitive at detecting whether something was present at all. But the electronic nose outperformed humans at recognizing which specific compound was present once concentrations were above the detection floor.¹² The researchers concluded that e-noses are best understood as complementary tools, not replacements. This makes intuitive sense: the human nose excels at detecting faint, complex odor qualities, while the e-nose excels at consistent classification of samples it has been trained to recognize.

In food quality control, e-noses have demonstrated their value precisely where speed matters. One study found that an electronic nose could detect guaiacol, a compound responsible for a medicinal off-flavor in apple juice caused by bacterial spoilage, with significantly higher sensitivity than a trained test panel.¹³ In peanut processing, an e-nose with 32 semiconducting polymer sensors could differentiate between peanut lots with and without off-flavor volatiles at a very high confidence level.¹⁴ For these binary pass/fail decisions (spoiled or not?), electronic noses often outperform panels because they deliver instant, repeatable results without panelist fatigue.

Medical Breath Analysis

One of the more surprising applications of odor measurement technology is in diagnosing disease. The idea is that metabolic changes caused by illness alter the volatile compounds in a patient’s exhaled breath, and an electronic nose can detect those changes. Gas chromatography studies have already demonstrated that breath analysis can reflect a person’s metabolic status, distinguishing normal from disease-associated profiles. Electronic noses have shown promising results in distinguishing patients with obstructive airway diseases, various cancers, and inflammatory and infectious conditions from healthy controls, with moderate to good accuracy.¹⁵

The appeal is obvious: e-nose analysis is portable, fast, inexpensive, non-invasive, and does not require a specialized technician. But the field is not yet ready for routine clinical use. A systematic review noted that standardized guidelines for breath sampling, analysis, and result interpretation are still lacking. Most existing studies have used relatively small populations, and external validation in independent patient groups is needed before the technology can be trusted for real diagnostic decisions.¹⁵ Reproducibility and repeatability of e-nose measurements also need further optimization. The science is genuinely exciting, but anyone selling breath-based diagnostics as proven medicine today is getting ahead of the evidence.

Biohybrid Sensors

The newest frontier in odor measurement borrows directly from biology. Biohybrid sensors combine actual biological olfactory components with electronic readout hardware. One approach embedded mosquito olfactory receptors into an artificial lipid membrane on a chip, creating a sensor that detected the volatile compound octenol by measuring ion current changes triggered by receptor activation.¹⁶ Another used the intact antennae of desert locusts, connected to electroantennogram electrodes and paired with artificial-intelligence signal analysis, to discriminate between different odors.¹⁷

The rationale is straightforward: biological olfactory systems have been refined by hundreds of millions of years of evolution to detect and classify volatiles with extraordinary sensitivity and selectivity. Rather than trying to engineer synthetic sensors that match that performance from scratch, biohybrid approaches leverage biology’s existing solutions while adding electronic readout and data processing. These systems are still firmly in the research phase. They face practical challenges around sensor lifespan (biological tissue degrades), environmental stability, and scalability. But they represent a genuinely different path from the MOS sensor arrays that dominate current e-nose technology.

Predicting Smell from Chemical Structure

A complementary computational approach skips the sensor entirely and tries to predict what a molecule will smell like from its chemical structure alone. Deep neural networks trained on a dataset of over 5,000 chemical compounds with 104 distinct smell labels achieved prediction accuracies above 97% on independent test sets.¹⁸ The models work from a text-based representation of molecular structure or from images of the chemical diagram, and they can predict multiple smell descriptors simultaneously for a single compound.

This kind of computational screening could accelerate fragrance and flavor development enormously, letting chemists identify candidate molecules with desired smell profiles before they are ever synthesized and sniffed. But the models share a fundamental limitation with electronic noses: they work with individual pure compounds and known training categories. Predicting the smell of a novel mixture with perceptual synergy and masking effects remains far beyond what current algorithms can handle.

Sampling Problems That Undermine Everything Else

Before any measurement method can do its job, someone has to collect and deliver the odorous air without changing it. This turns out to be harder than it sounds. Many odorants are chemically reactive or present at trace concentrations, and they interact with container walls, tubing, and sorbent materials. A study evaluating field sampling methods for humid environments found that Tenax tubes, a common sorbent material, suffered from breakthrough of volatile compounds and tended to form chemical artifacts as sampling volumes increased.¹⁹ The result is that the sample reaching the laboratory or the e-nose does not always faithfully represent what was present at the source.

Humidity is a particularly persistent problem. Water vapor interferes with metal oxide sensors, can displace odorants from sorbent materials during sampling, and changes how volatile compounds partition between the gas phase and surfaces. In hot, humid environments like wastewater treatment plants or tropical food-processing facilities, getting a reliable sample can be the hardest part of the entire measurement process. Researchers have developed specialized sampling cartridges and drying systems to address this, but no single approach works perfectly for all odorant classes. Anyone interpreting odor data should ask how the sample was collected and handled before trusting the numbers, because a perfectly calibrated instrument analyzing a degraded sample produces perfectly calibrated nonsense.

1
Atmospheric Environment. Odor measurements according to EN 13725: A statistical analysis of variance components
2
PubMed. Relevance of n-butanol as a reference gas for odorants and complex odors
3
PubMed. Global study of variability in olfactory sensitivity
4
PubMed Central. Recent odor experience selectively modulates olfactory sensitivity across the glomerular output in the mouse olfactory bulb
5
PubMed. Development of the Relationship between Odor Intensity and Threshold Dilution Ratio for Swine Units
6
Chemical Senses. Synergy and Masking in Odor Mixtures: An Electrophysiological Study of Orthonasal vs. Retronasal Perception
7
PubMed Central. Odour Detection Methods: Olfactometry and Chemical Sensors
8
PubMed. GC-O-MS technique and its applications in food flavor analysis
9
PubMed Central. A Comparative Analysis of Cold Brew Coffee Aroma Using the Gas Chromatography-Olfactometry-Mass Spectrometry Technique
10
PubMed Central. Metal oxide sensors for electronic noses and their application to food analysis
11
Engineering Applications of Artificial Intelligence. Metal oxide semiconductor based electronic nose data pre-processing, review
12
PubMed. Evaluation of wine aromatic compounds by a sensory human panel and an electronic nose
13
Food Control. Detection of medicinal off-flavor in apple juice with artificial sensing system and comparison with test panel evaluation and GC–MS
14
Transactions of the ASAE. A Method to Detect Peanut Off–Flavors Using an Electronic Nose
15
PubMed Central. The electronic nose technology in clinical diagnosis: A systematic review
16
PubMed. Construction of a Biohybrid Odorant Sensor Using Biological Olfactory Receptors Embedded into Bilayer Lipid Membrane on a Chip
17
PubMed. The Locust antenna as an odor discriminator
18
PubMed. SMILES to Smell: Decoding the Structure-Odor Relationship of Chemical Compounds Using the Deep Neural Network Approach
19
PubMed. Field sampling method for quantifying odorants in humid environments