In 2025 I bought a decaf lot out of Ethiopia. The sample cupped well and carried a blueberry note, the kind you get from Ethiopians, and I ordered a large batch on it. What arrived tasted dramatically different, and it was difficult to roast besides; I was re-cutting the profile about once a month for as long as we had it in stock.
Here is what I actually got wrong. I treated the smell as the most reliable thing in that sample, and it is the least reliable thing in a coffee. Almost none of it exists in the green seed. All of it is assembled during roasting, out of raw material that on its own smells of very little: protein, a small pool of loose sugar, a plant acid, and a pigment. And a meaningful part of it starts leaving within hours of the roast.
Coffee smells like coffee because roasting builds the smell from scratch, and it does not build it with one reaction. There are at least four separate chemical routes running at the same time, on four different ingredients, and they are not variations of each other. That is the part I find genuinely useful, and it took reading the food-science literature rather than the coffee literature to see it.
Does green coffee smell like coffee?
No, and the gap is larger than most people expect. One review counts around 300 volatile organic compounds in the raw seed against more than 850 in the roasted one. Worse, some of what the green seed does smell of is the thing roasting has to destroy: raw coffee carries what the literature calls peasy off-odors, from a family of methoxypyrazines including 3-isobutyl-2-methoxypyrazine, the same class of compound that makes an under-ripe bell pepper smell green.
This part is not ours and I am not going to present it as a finding. Barista Hustle says it cleanly: "The smell of green coffee beans bears little similarity to the smell of roasted coffee." Perfect Daily Grind puts it the same way. Both are worth reading and both stop at the level of precursor classes, which is exactly where the interesting part begins.
What the green seed holds is a shopping list, not a smell. Protein at roughly 11 to 15 percent by dry weight. A small free pool of amino acids floating loose. Loose glucose and fructose, plus sucrose as a reservoir. Chlorogenic acids at 5 to 8 percent. And carotenoid pigments, in small quantities. Four of those five go on to make aroma by routes that have nothing to do with one another.
Which single compound smells most like coffee?
2-furfurylthiol, and it is not close. In the aroma-extract dilution work that Werner Grosch's group ran on roasted arabica, its recorded odor quality is simply "roasty (coffee)." Later reviews describe it as the fresh-coffee note itself. If you could smell one molecule out of the thousand-plus that have been identified in roasted coffee and call it coffee, this is the one.
Its detection threshold in air is somewhere around 10 to 20 picograms per liter. A picogram is a trillionth of a gram. You can get to that concentration from a quantity of material you would struggle to weigh.
The route to it is the part worth knowing, because it explains why the note is fragile. Cysteine, one of the sulfur-bearing amino acids, breaks down under heat and releases hydrogen sulfide. Separately, pentose sugars degrade into furfural and then furfuryl alcohol. The two halves meet, lose a water molecule, and you have furfurylthiol. Work on which sugar feeds the reaction best puts ribose first, ahead of xylose, fructose and glucose.
Two ingredients that individually smell of rotten egg and burnt sugar make the smell people describe as the best part of the morning. That is not a metaphor for anything. It is the actual mechanism.
Which amino acid makes which smell?
This is the mapping I went looking for and could not find anywhere in coffee writing, so here it is as plainly as the sources support.
Strecker degradation is a side branch of the Maillard reaction in which an amino acid meets one of the reaction's two-carbonyl intermediates, loses a carbon as carbon dioxide, and becomes an aldehyde one carbon shorter than it started. We have written about the carbon-dioxide half of that before, because it is where second crack and the bloom in your brewer come from. The aldehyde half is the smell.
Each amino acid gives its own aldehyde, and each aldehyde has its own odor:
- Leucine becomes 3-methylbutanal, recorded in the coffee aroma tables as malty.
- Isoleucine becomes 2-methylbutanal, described in the Maillard literature as malty and chocolate.
- Valine becomes 2-methylpropanal. In espresso work the three branched-chain aldehydes are treated as a group and read as malty and fermented.
- Methionine becomes methional, potato-like and sweet.
- Phenylalanine becomes phenylacetaldehyde, floral or honey-like.
Two honesty flags on that list, because it would be easy to present it as tidier than its sourcing. I verified the amino-acid-to-aldehyde pairings in flavor-science reviews rather than in a coffee paper, since the coffee work I could read names the aldehydes and their odors without tracing each one home. And methional is not universally welcome in coffee: Seninde and Chambers' review lists the potato note among the negative flavor compounds.
There is also a live disagreement worth naming rather than smoothing. Perfect Daily Grind's aroma piece, which is the best general-audience treatment in coffee writing, describes 3-methylbutanal as creating "fruity and sweet aromas." Grosch's dilution work records it as malty. I have gone with malty because that is what the primary measurement says, but the two descriptions are not reconcilable and you should know they both exist.
Why does a roast smell nutty and roasty rather than malty?
Because the Strecker aldehydes are not the end of it. The same reaction produces aminoketones, and those condense with each other into pyrazines, which are a different class of compound with a different smell. Pyrazines are the second most abundant volatile group in roasted coffee, at something like 25 to 39 percent of the total, and they are where the nutty, roasty, earthy character lives.
Two of them carry unusual weight. 2-ethyl-3,5-dimethylpyrazine came out of Grosch's dilution work at the joint highest impact score in the whole extract, with a threshold in air of about 7 to 14 picograms per liter. 2,3-diethyl-5-methylpyrazine sits beside it. Both are recorded as "earthy, roasty."
So one starting material, the free amino acid pool, feeds two different smells by two different exits, and which exit dominates depends on the temperature and the time. That is a roasting lever. It is not a precise one.
Where do the smoky and fruity notes come from, then?
Not from the amino acids at all, and this is the point at which the usual explanation stops being sufficient.
The phenolic notes come from chlorogenic acid. Guaiacol, 4-vinylguaiacol and 4-ethylguaiacol are thermal breakdown products of the chlorogenic acids, by way of ferulic acid, and their recorded odors run phenolic, spicy and burnt. 4-ethylguaiacol's threshold in air is around 10 to 30 picograms per liter; 4-vinylguaiacol's is 0.4 to 0.8 nanograms per liter. Chlorogenic acid was sitting in the green seed at 5 to 8 percent of dry weight the whole time, doing nothing for the smell until heat took it apart.
And then there is the one that surprised me most. β-damascenone has the lowest odor threshold of any compound in these tables, around 2 to 4 picograms per liter in air, and it tied for the highest impact score in Grosch's extract. It is described as honey-like and fruity. It is not a Maillard product at all. It comes from carotenoids breaking down, which is to say it comes from the seed's pigment rather than from its protein or its sugar.
I want to be straight about how thin that last claim is in the sources I could read. The carotenoid origin appears in the coffee literature in about five words, in one review, plus a generic precursor list in another. The high impact and the low threshold are very well evidenced. The origin, in coffee specifically, is asserted rather than demonstrated in anything I was able to open, and the Poisson and Hofmann chapter that would settle it sits behind a paywall I could not get through. Treat it as likely rather than as established. Other reviews also describe β-damascenone as creamy and caramel-like rather than honeyed, so even the smell is not agreed.
There are five routes on that list, running on four different starting materials: amino acids through Strecker, amino acids again through the pyrazines, cysteine plus a pentose through the sulfur route, chlorogenic acid through the phenolics, and carotenoids through β-damascenone. That is the actual architecture of coffee aroma, and there is no single dial on a roaster that moves all of them together.
Why does coffee stop smelling like that?
Faster than almost anyone plans for, and the sulfur compounds go first.
The Specialty Coffee Association's own literature review on coffee staling is the best practitioner source here and I would rather send you to it than restate it. Its numbers on methanethiol, another sulfur volatile, are the ones that stuck with me: it "dissipated two hours after roasting," and after eight days of storage it had fallen to about 30 percent of its original amount. The review also puts the greatest rate of freshness loss in the first month.
Furfurylthiol goes the same direction, and the mechanism is worth a sentence because it is unexpectedly elegant. It does not simply evaporate. It bonds covalently to melanoidins, the brown polymers the Maillard reaction leaves behind, so the compound that makes coffee smell like coffee is captured by the material that makes coffee brown. Oxygen accelerates it: work cited across the packaging literature found the drop was rapid at oxygen levels at or above 5 percent and much slower around 2 percent.
I am deliberately not giving you a half-life for furfurylthiol. I looked for one, the papers that would have it were behind paywalls, and a number I cannot open the source for is not a number I will publish. What I can tell you is the direction, the mechanism and the oxygen dependence. If you want the practical version, it is the same advice as always about keeping air away from roasted coffee, with a better reason attached.
What does any of this change about the bag on your counter?
Three things, and the first is a concession.
We publish tasting notes on our single origins and a cup score on everything, and neither is an aroma measurement. Tasting notes describe a cup somebody drank. They do not tell you which of these compounds the roast built or how much of them is left by the time the bag reaches you. That is not something we publish, and I have not found it published anywhere I could check, which is uncomfortable to write in an article where I am also the person who bought a lot on an aroma and got it wrong. If what you are buying is the smell, buying coffee by mail is a bad way to do it, because you cannot smell a bag through a screen and the part you care about is the part that leaves first.
What we do about it is structural rather than clever. We roast once a week, on Wednesdays, and our own shop copy says coffee ships within two business days of the roast rather than within two business days of your order. That difference is the cost of a weekly roast and we carry it on purpose: it means we cannot smooth demand across the week the way continuous roasting would, and some orders wait on a roast day. What it buys is that the coffee leaves here close to the roast instead of sitting in a warehouse, which for the sulfur compounds is the whole argument.
Second, processing matters to aroma before roasting ever starts. Fermentation builds esters in the seed, and natural processing in particular loads the green coffee with fruit-side aromatics that are not Maillard products and are not on the list above. Our own processing-method routes cover that side at length. This article has stayed on the roast-derived half deliberately, because they are genuinely different chemistry with genuinely different levers.
Third, and back to the decaf. I still do not know how much of that lot's change was the coffee and how much was me. The entry in my own notes says the profile needed adjusting about monthly and that it was difficult to roast, and both of those are true, and neither of them tells me whether I was chasing a moving coffee or failing to hit a fixed one. Decaf is harder to roast for reasons we have written up separately, so there is a real explanation available. I am not confident enough to claim it. What I took away instead is narrower and I have stuck to it: a sample tells you what a coffee did on one day in one roaster, and the aroma is the part of that reading with the shortest shelf life.
Where do these numbers come from?
The measurement that underpins most of this is Imre Blank, Ashok Sen and Werner Grosch's aroma-impact work on arabica and robusta, presented at the 14th ASIC colloquium in 1991, which is where the odor descriptors and the impact rankings come from. The compound counts, thresholds and formation routes are drawn from Zakidou and colleagues in Molecules (2021), Angeloni and colleagues in Molecules (2021) on espresso key odorants, Cao and colleagues' 2023 review in the International Journal of Food Science and Technology, Gao and colleagues in Frontiers in Nutrition (2022) on Maillard flavor products, and Zhang and colleagues in Frontiers in Nutrition (2023) on furfurylthiol. Seninde and Chambers' 2020 review in Beverages is the best open-access overview of coffee flavor generally. Barista Hustle on Strecker degradation and Perfect Daily Grind on aroma chemistry are where I would send a reader next in coffee writing specifically.
What is ours is the part you cannot get from a paper. We roast this coffee, we pick the pale seeds out of the cooling tray by hand, we write the tasting notes, and we are the ones who have to decide whether a sample is telling the truth. I got that call wrong in 2025, and it cost me a re-cut profile about every month for as long as that coffee was on the rack.
Which coffee on our shelf smells most like this?
Ethiopia Buku Abel is the most aromatic coffee on our shelf right now, a natural-process lot from the Guji Zone that cups at 93, the highest score we currently have. Our notes on it are mango, papaya, berries, chocolate and chai spice. Open the bag before you grind it and you will get most of that off the dry grounds, which is the fastest demonstration of everything above that I know of. We roast on Wednesdays and the cup score is on its page.
