The Evolution of Scent: What 170,000 Perfumes Reveal About How Culture Changes

The Science

In the spring of 1921, Ernest Beaux handed Coco Chanel a small glass vial. The liquid inside was unlike anything that had come before it—aldehydic, abstract, impossibly modern. Chanel named it No. 5, released it that May, and fundamentally altered what perfume could be. A century later, researchers at Yale, MIT, and the Max Planck Institute are asking a question that would have seemed absurd to Beaux: what can a database of 170,000 perfumes tell us about how culture evolves?

The answer, it turns out, is quite a lot.

Vahid Satarifard and colleagues have assembled what they believe to be the largest perfume corpus ever studied—a digital monument to a century of human attempts to smell interesting. Drawing from three independent fragrance databases (Fragrantica, Parfumo, and Aromo), they compiled 169,759 perfumes, 8,332 brands, 3,050 individual perfumers, and approximately 25.9 million crowd-sourced votes on everything from longevity to likability. The temporal window spans from 1900 to 2024, capturing the full arc of modern perfumery from the golden age of French houses to the current era of niche brands and gender-neutral marketing.

The methodological ambition here deserves attention. This isn't simply a historical survey or a market analysis. The researchers have constructed what they call a "multidisciplinary computational framework" that borrows tools from information theory, population genetics, and network science. They treat perfumes as a cultural population—subject to selection pressures, driven by imitation, and evolving in ways that can be quantified.

To characterize individual perfumes, they deployed two complementary measures. Complexity, defined as the Shannon entropy of a perfume's note distribution, captures how many different notes a fragrance contains and how evenly they're distributed. A perfume with ten different notes, each contributing equally, has higher complexity than one dominated by a single accord. Novelty, measured as the co-occurrence surprisal of a perfume's note combinations relative to all earlier perfumes in the corpus, captures how unexpected a fragrance's particular blend is. A novel perfume doesn't just use rare ingredients—it uses them in combinations that haven't been seen before. (Satarifard et al., 2026)

Perhaps the most innovative methodological move involves something perfume enthusiasts call "flankers" and "clones." A flanker is a spin-off of a successful perfume from the same brand—think of Dior Homme and Dior Homme Intense, or Chanel Egoiste and its descendants. A clone is a close copy of another brand's fragrance, made by a different house. What makes these categories scientifically valuable is that they're explicitly annotated on Fragrantica by crowd-sourced users: if thousands of people agree that Perfume X resembles Perfume Y, that's a verifiable signal of cultural transmission.

The researchers used these resemblance annotations to construct time-directed copy-lineage networks—essentially family trees for fragrances. If perfume B resembled perfume A and came later, B became A's offspring in this network. By iterating backward from any given perfume, they could identify its "parents" in a cultural sense. They found 169,991 highly-voted resemblance pairs in their dataset, with a mean Jaccard index of shared notes at 0.17—substantially higher than the 0.09 you'd expect by chance (Satarifard et al., 2026).

Finally, they mapped the collaboration network of master perfumers—the web of creative partnerships that connects people who have worked together on formulations. By examining how the stylistic signatures of one perfumer appeared (or didn't appear) in the work of their collaborators, they could test whether creative style behaves as a social contagion that spreads through professional networks.

The cross-database validation is worth noting. The researchers deliberately drew from three independent sources and showed close agreement on composition, release history, and note usage. Fragrantica provided the richest data due to its larger user base, but Parfumo and Aromo served as validation datasets, reducing the risk that patterns reflected the idiosyncrasies of a single platform.

What They Found

The first major finding concerns a paradox that emerges from the history of perfume composition over the past 125 years: fragrances have become simultaneously simpler and stranger.

Complexity—that Shannon entropy measurement—showed a clear and dramatic decline beginning in the early 1990s. From roughly 1965 to 1990, the average perfume had a complexity score of about 3.7 bits, indicating a relatively even distribution across multiple notes. After 1990, this dropped abruptly to around 3.0 bits and remained stable through 2024. The researchers observe the same pattern using a uniform-weight variant of complexity in the Parfumo dataset, strengthening confidence in the result.

But here's what makes this interesting: at the same time perfumes were becoming structurally simpler—fewer notes, more dominance by a single accord—they were also becoming more novel. Note-pair novelty climbed from about 2.3 bits in the early 1900s to 7.6 bits in the 2020s, with the steepest gains occurring from the 1990s onward. The researchers validated this trend using an independent semantics-based measure (Word2Vec embeddings of notes) that doesn't rely on raw combinatorial counts, ruling out the possibility that rising novelty was simply an artifact of an expanding ingredient palette.

The tension between these two trends becomes visible when you plot complexity against novelty for individual perfumes. The Spearman correlation is small but significant (ρ = -0.13 for pairs, -0.15 for triples): more complex perfumes tend to be less novel, and vice versa. The researchers trace a decade-averaged trajectory through this space that moves steadily from the upper-left (complex, conventional) toward the lower-right (simple, unconventional). This trajectory passes near some iconic examples: the high-complexity, low-novelty frontier is occupied by classics like Chanel No. 5 (1921), Guerlain's L'Heure Bleue (1912), and Lanvin's Arpège (1927)—fragrances built on well-established notes in traditional proportions. The low-complexity, high-novelty frontier features more recent releases like Wood Sage & Sea Salt (2014), Baccarat Rouge 540 (2015), and Molecule 01 + Mandarin (2021)—minimalist compositions that combine notes in ways that hadn't been tried before (Satarifard et al., 2026).

This structural shift seems to align with changes in consumer preferences. The researchers regressed user ratings on complexity, novelty, longevity, sillage, and price value for both the full dataset and a contemporary subset (2000-2024). Longevity had by far the strongest positive association with ratings: people like perfumes that last. But novelty showed a significant negative association in both groups, and complexity became significantly negative in the contemporary subset. Users, it seems, prefer simpler and less experimental fragrances—which makes the rise of novelty-driven minimalist perfumery something of a creative risk taken despite rather than because of consumer preference.

Rising Novelty in Perfume Composition

Label	Value
1900s (early)	2.3
1920s	3.1
1940s	3.4
1960s	4.1
1980s	4.8
1990s	5.3
2000s	6
2010s	6.8

Source: Satarifard et al., 2026, Figure 2B. Novelty measures co-occurrence surprisal of note pairs and triples against all earlier perfumes.

The second major finding concerns cultural selection operating through imitation. The copy-lineage networks revealed something striking: while imitation is pervasive in perfumery, original creations retain a measurable quality premium. Originals (perfumes with no identified parent in the lineage network) had significantly higher longevity, sillage, and user ratings than either flankers (brand extensions of existing perfumes) or clones (copies from other houses). The effect sizes were small but consistent: originals outperformed flankers and clones on all three metrics, with the largest difference appearing in sillage (Cohen's d = 0.23 for originals vs. clones) (Satarifard et al., 2026).

This matters because it suggests that the cultural selection operating on perfumes isn't purely imitative. Copying successful formulas is common—the dataset contains 12,028 clones and 2,222 flankers compared to only 123 identified originals—but original creations still tend to have better performance characteristics. The Price equation, borrowed from evolutionary biology, allowed the researchers to decompose this pattern: they found that an increasing share of notes "drift free" of selection over time, meaning that their usage isn't strongly predicted by their copying dynamics. Originality and quality appear connected, even in a market saturated with variations on established themes.

The two largest genealogical lineages in the data illustrate the scale of imitative dynamics. Chanel No. 5 seeds lineage to 68 perfumes spanning nearly a century—flankers, inspired interpretations, and occasional clones that carry its genetic signature forward. But even more dramatically, Creed's Aventus (released in 2010) has generated 36 descendants in just over a decade, making it one of the most influential masculine fragrances of the modern era. The researchers note that the CD index—a measure of cultural descendance—has remained stable over the long term but shows interesting variation across eras. In the 1985-2004 period, over-copying and rising use were negatively coupled: notes that were copied more than average actually tended to decline in prevalence. This is a signature of cultural selection, where success in transmission doesn't guarantee success in the population.

Quality Premium for Original Perfumes

Label	Value
Originals	3.65
Flankers	3.45
Clones	3.35

Source: Satarifard et al., 2026, Figure 3B. Error bars represent 95% confidence intervals. Effect sizes (Cohen's d) reported for originals vs clones on sillage.

The third major finding concerns the social transmission of creative style among perfumers. The collaboration network of master perfumers—mapped from their joint work on formulations—reveals patterns consistent with social contagion. The annual count of collaborative perfumes has grown substantially over time, rising from roughly 20% of all releases in the early 1900s to over 60% by 2020. This reflects the industrial transformation of perfumery, where houses increasingly pool creative talent rather than relying on individual noses.

But the more interesting question is whether style itself spreads through these networks. The researchers tested this by examining whether a perfumer was more likely to adopt certain note or accord descriptors from a collaborator with whom they shared a direct network connection, versus a randomly sampled perfumer at the same social distance. They found a significant effect: the probability of stylistic adoption decreased with network distance but remained elevated even at three degrees of separation. Creative style, it seems, behaves like a contagion—transmitted through direct collaboration but decaying as it moves through the professional network (Satarifard et al., 2026).

The 3-core of the largest collaboration cluster—essentially the densest subgraph of the network—includes many of the industry's most prolific figures. Alberto Morillas, the most productive master perfumer with nearly 500 commercial fragrances to his name, appears as a central node connecting multiple communities. Standardized regression coefficients show that shared brand affiliation, era proximity, and note similarity are all significant positive predictors of collaboration, but accord similarity and same-country origin are not—which suggests that collaborative relationships form around brands and time periods rather than purely stylistic affinity.

Perhaps most fascinating is the relationship between network position and creative style. Perfumers with higher eigenvector centrality (meaning they're well-connected to other well-connected perfumers) tend to produce perfumes with lower complexity and higher novelty. This could suggest that centrality exposes perfumers to a wider range of influences, pushing them toward more experimental territory. Or it could reflect selection: perfumers who produce simpler, more novel work may be more attractive collaborators. The data can't cleanly distinguish these mechanisms, but the pattern is robust.

The researchers also examined the "CD5 index"—a measure of how much a perfumer's stylistic descendants resemble them over a five-year window—for those who have ever collaborated versus those who always work alone. Collaborators show lower CD5 scores on average, consistent with the idea that working with others dilutes stylistic distinctiveness over time. Style is transmitted, but it mutates in the process.

Declining Complexity in Modern Perfumery

Label	Value
1965-1990	3.7
2000-2024	3

Source: Satarifard et al., 2026, Figure 2A. Complexity measures the Shannon entropy of a perfume's note distribution.

Beyond these core findings, the paper documents several additional patterns worth noting. The global geography of perfume reveals striking regional signatures: Western Europe produces roughly half of all catalogued perfumes, with the Anglosphere adding another quarter. But accord profiles cluster by region in ways that reflect longstanding cultural traditions. Middle Eastern countries over-index on oud, amber, and warm-spicy accords—consistent with oriental perfume traditions that predate modern commercial perfumery by centuries. Latin America gravitates toward sweet and fruity combinations, while the Anglo-West prefers fresh and green notes. Region explains about 13% of the variance in note profiles and 17% in accord profiles, a modest but significant effect.

The temporal evolution of gendered marketing tells a story of gradual liberalization. Female-marketed perfumes dominated for most of the 20th century, but since the late 1990s, unisex fragrances have risen sharply—a trend aligned with broader cultural movements toward fluid identity expression. The gender coding of individual notes and accords has shifted, but not uniformly toward neutrality. Among notes that changed significantly, about 61% moved toward masculine coding rather than neutral. Incense, spicy notes, and patchouli crossed the neutral line in the masculine direction; tuberose, iris, and jasmine remained firmly feminine.

The macro-level release dynamics reveal three distinct eras separated by inflection points around 1918 and 2014-2018. The first era, which the researchers call Classic Perfumery, saw gradual growth up to about 20 releases per year. A Stagnation period followed, spanning roughly 1926 to 1979, where growth plateaued. Then the Industrial Perfumery era began, accelerating toward a carrying capacity of roughly 6,000 new perfumes per year—an explosion that the researchers connect to industry consolidation (the major houses absorbed dozens of independents through the 1980s and 1990s) and to the environmental traces detectable in Caucasus ice cores.

Why This Changes Things

The central claim of this paper is one that should expand how we think about cultural evolution. Satarifard and colleagues argue that perfume constitutes a "high-resolution model system" for studying how sensory culture changes over time—a system with properties that make it unusually tractable for quantitative analysis.

Consider what makes perfume tractable. In music, the object of cultural transmission (a song) is a complex, multidimensional object with disputed boundaries. In language, change operates across multiple levels—phonological, syntactic, semantic—and the signals are noisy. In fashion, the units of cultural evolution (garments, styles, aesthetics) resist clean discretization. But perfume has the crucial advantage that each fragrance can be decomposed into a finite palette of discrete notes—jasmine, sandalwood, bergamot, musk—and the combinations are enumerable. The crowd-sourced annotation of resemblance means you can construct actual lineage networks, not just infer them from stylistic similarity. The performance metrics (longevity, sillage, ratings) provide outcome measures for selection.

This makes perfume arguably the cleanest cultural evolution system outside of genetics. The Price equation—the same mathematical framework used to decompose genetic evolution into selection and drift components—applies directly. You can quantify the fitness of a cultural variant (how many times it's copied), measure the selection differential (how much copying predicts use), and partition the change in a population into components attributable to selection versus drift. These are not metaphors; they're direct applications of evolutionary theory to cultural data.

The finding that original creations retain a quality premium is particularly significant for theories of cultural evolution. Standard models often treat cultural transmission as essentially imitative: variants spread because they're copied, not because they're better. But the perfume data suggests that quality still matters. Originals don't just have more "cultural offspring" on average; they also have measurably better performance. This hints at a role for innovation that pure copying models struggle to accommodate. The cultural selection operating in perfume is neither purely blind (copying without regard to quality) nor purely rational (spreading only the best variants). It's a hybrid: imitation is pervasive, but original quality leaves a trace.

The complexity-novelty tradeoff is also theoretically significant. The researchers frame this as a case of "exploratory variation under selection constraints": perfumers are pushed toward simpler compositions by consumer preferences (which favor lower novelty) and regulatory pressures (which restrict certain ingredients), but they're pulled toward more novel combinations by competitive dynamics (where surprise can capture attention in a crowded market). The result is a population that simultaneously simplifies and radicalizes—a pattern that may characterize other creative domains.

The social contagion finding extends this to the individual level. Style, it seems, is not purely a matter of individual talent or aesthetic vision. It's something that flows through professional networks, shaped by who you've worked with and how closely. A perfumer's creative signature—what makes their work recognizable—partially reflects the accumulated influence of collaborators, diluted across degrees of separation. This has implications for how we understand creative authorship and the evolution of artistic traditions: the nose doesn't work alone.

The practical implications for the industry are also worth considering. The researchers' findings suggest that the current moment—minimalist fragrances with rare note combinations—represents an unstable equilibrium. User ratings consistently favor simpler, less novel compositions, but the competitive dynamics of a saturated market reward novelty. This creates pressure toward the experimental frontier even as consumer satisfaction lags. Whether this tension resolves through market consolidation, consumer education, or a taste correction toward the classics remains to be seen.

For the broader public, the significance is more conceptual. Perfume is one of the oldest human technologies for modifying our chemical signatures—our social presentation through scent. That this practice has been quantitatively mapped across a century of cultural evolution should matter to anyone interested in how human culture actually works, as opposed to how we theorize it. The perfume corpus is a record of collective aesthetic decision-making at extraordinary resolution: every fragrance represents thousands of choices, made by perfumers and consumers, about what should smell good. To see the patterns in those choices is to see culture in the making.

What's Next

Several open questions emerge from this work, and the researchers are honest about them.

First, causality remains elusive. The correlation between originality and quality could reflect several mechanisms: originals may be better because they're genuinely innovative (a quality effect), or because brands invest more in originals than flankers and clones (a resource effect), or because the very act of being recognized as an original grants status that translates into favorable ratings (a halo effect). The data can distinguish these possibilities imperfectly at best. Experimental or quasi-experimental designs—comparing flankers that were promoted aggressively versus those that weren't, or tracking quality metrics before and after a brand's original becomes a cultural reference point—could help disentangle the mechanisms.

Second, the social contagion analysis is limited to the 20th century at best, and the collaboration network density increases sharply in recent decades. It's possible that the contagion dynamics described here operate differently in earlier periods, when the industry was more fragmented and perfumers worked more independently. The researchers acknowledge this and present their collaboration findings as preliminary. Reconstructing historical collaboration networks from archival sources—house records, biographies, trade publications—could extend the analysis and test whether the contagion patterns hold across different organizational structures.

Third, the regulatory story deserves more attention. The researchers note that the International Fragrance Association (IFRA) has progressively restricted ingredients like oakmoss, a known skin sensitizer, and that this has driven compositional change. But they don't systematically quantify how much of the complexity decline is attributable to regulatory pressure versus other factors (consumer preference shifts, market consolidation, changes in fashion). A dedicated analysis of regulatory events and their compositional correlates could identify specific causal pathways.

The geographic analysis, while suggestive, also needs refinement. The finding that region explains 13-17% of variance in accord profiles is modest but significant. But the dataset is biased toward Western catalogues—Fragrantica, in particular, has strong European and North American coverage but thinner data from Asia, Africa, and South America. The Middle Eastern signature is clear because Middle Eastern perfumery is well-documented in the databases, but other regions may have distinctive patterns that aren't captured. Expanding the corpus to include regional databases with better local coverage would strengthen the geographic claims.

Finally, there's the question of what perfume evolution might predict. If cultural evolution in perfume follows patterns that also characterize other sensory domains—music, cuisine, fashion—then understanding perfume could illuminate those other domains. The researchers speculate that perfume might serve as a "canary in the coal mine" for cultural dynamics that operate more slowly elsewhere, because its short product cycles and clear unit boundaries make it easier to track. Whether this promise is realized depends on whether the structural patterns identified here—complexity-novelty tradeoff, quality premium for originals, social contagion of style—replicate in domains where the data is messier and the unit boundaries fuzzier.

The researchers conclude by positioning perfume as "a culturally evolving system, akin to music and fashion, through which societies express and communicate hedonic sensory experiences." This framing deserves emphasis. The study of cultural evolution has often focused on domains where the data is conveniently digitized—books, music, patents—because that's where computational analysis is tractable. But smell is different: it's arguably the most ancient and most visceral of our senses, operating through neural pathways that predate language. To see the same evolutionary dynamics operating in perfume that we see in music and language is to suggest that these dynamics are universal features of cultural systems, not domain-specific artifacts.

That implication extends beyond academia. Every perfume ever made is a hypothesis about what should smell good—a bet on collective aesthetic preferences that may or may not pay off. The fact that we can now trace the outcomes of those bets across a century, and see patterns in what worked and what didn't, is a testament to the power of large-scale cultural data. The perfume industry has always understood that scent is a form of communication. This paper suggests it may also be a form of evolution—subject to the same forces of variation, selection, and drift that shape all living systems.

The next time you catch a whiff of something familiar and feel suddenly transported to another time or place, remember: that association didn't happen by accident. It accumulated across decades of cultural choice, encoded in a bottle, transmitted through imitation and innovation, shaped by the invisible hand of collective preference. The perfume knows more about culture than we usually admit. Now, thanks to this research, we can begin to read what it knows.