People smelling the same fragrance material may describe it as “floral,” “sweet,” or “reminiscent of detergent.” Experts’ rich vocabulary helps describe scent characteristics, but what people without regular olfactory training perceive needs to be studied separately.
In 2025, Antonie Louise Bierling and colleagues published a large open dataset of responses from laypeople who actually smelled the stimuli. It covered 74 monomolecular odors overall, with 1,119 participants in the main healthy sample. The researchers recorded nonexperts’ perceptions using both freely chosen words and common rating scales.
Recording laypeople’s perceptions too
The starting point was the relationship between molecular properties and perceived odor. Why does a molecule smell sweet, and how pleasant does it seem? Exploring such relationships requires not only information about molecules but also data on how people actually perceive them.
The authors identified a problem: much existing odor data relies on trained experts’ assessments. Expert judgments are valuable, but a single description from a small group cannot capture differences among laypeople. Offering response options from the outset may also prompt people to think about a scent through those words.
Participants therefore first described each odor without response options, then evaluated it on shared scales. Free descriptions preserve impressions that come to mind spontaneously; common scales allow comparisons between people and molecules. The combination gives the dataset its value.
Each participant smelled ten of the 74 odors
In the main study, each person received ten odors: two common to everyone and eight that varied between groups. Dividing the stimuli into nine sets reduced individual burden while covering 74 odors across the study. Not all 1,119 participants smelled all 74 odors.
Among the healthy participants, 875 evaluated odors at home and 244 in the laboratory. The paper additionally included 48 patients with reduced olfactory function and a separate group of 60 healthy participants for retesting, bringing the reported overall sample to 1,227. The 60 retest participants evaluated six odors twice, about a week apart.
After free description, participants rated eight dimensions: pleasantness, familiarity, intensity, irritation, edibility, disgust, warmth, and coolness. They also indicated whether 16 descriptors, including sweet, fruity, floral, woody, and musky, applied. The study was conducted in German.
The two common odors also provided reference points for comparing groups. If every group smelled entirely different odors, it would be difficult to tell whether rating differences arose from the odors, participants, or measurement conditions.
Pleasantness was relatively stable, while intensity was more sensitive to conditions
This paper was not a trial testing whether scent improves an outcome. It was a Data Descriptor reporting how data were produced and evaluating their reliability. Key results concern how closely the ratings corresponded to another study or to measurements on another day.
Across 56 odors shared with an earlier lay-participant dataset, pleasantness rankings correlated at 0.68. Familiarity, intensity, and some other dimensions also showed positive associations, but warmth and coolness did not correspond clearly. Agreement between studies therefore depends on which aspect of the impression is measured.
Retesting the same participants about a week later also revealed differences between dimensions. Across the six odors, correlations were 0.81 for pleasantness and 0.63 for intensity. Pleasantness showed relatively strong stability, while intensity varied more. A correlation of 0.81 does not mean that 81% of people gave the same answer; it describes how closely the ordering of ratings corresponded across measurements.
Comparing the two common odors between home and laboratory settings, participants at home rated them as slightly weaker. There were no clear differences in pleasantness and some other measures, so measurement setting did not affect every scale in the same way.
A large sample does not produce one “correct” perception
The dataset provides a basis for examining both average impressions and variation in responses. Looking at how many people selected a descriptor, for example, can capture a wider range of perceptions than fixing one molecule to one word.
However, participants tended to be young and highly educated, and German-language findings cannot simply be applied to Japanese speakers or a broad range of ages. Home and laboratory samples differed in participants, odor sets, and recruitment periods. We cannot attribute weaker home ratings solely to location.
Moreover, these were ratings of monomolecular odors, not direct predictions of preference for complete perfumes containing multiple fragrance materials. Whether impressions remain the same when concentrations or combinations change needs separate testing.
Summary
The study established a basis for investigating laypeople’s odor perceptions through actual ratings from a large sample. Combining free descriptions with common scales preserved both spontaneously chosen words and ratings that can be compared across people.
Validation showed some stability in dimensions such as pleasantness, alongside differences in agreement depending on what was evaluated and under which conditions. This foundation allows more detailed investigation of which molecular properties relate to perception and how those relationships vary between people. Rather than assigning one description to a scent, it provides material for connecting molecular properties with the range of sensations experienced by many people.
References
Bierling, A. L., Croy, A., Jesgarzewsky, T., et al. (2025). A dataset of laymen olfactory perception for 74 mono-molecular odors. Scientific Data, 12, 347.
DOI: 10.1038/s41597-025-04644-2
Full original article: Open full text at Scientific Data
Rights: The original article is published under CC BY 4.0.
