Beginner perfumer #107: how much of each material do you actually use? An answer measured from 954 formulas

About a 5 minute read.

The short version

• 187 materials appear in 20 or more of the 954 formulas; I computed each one's median share of the formula
• The overall median is 2.11%. Highest 16.16% (hexyl cinnamaldehyde), lowest 0.20% (Ambroxan), a spread of 81-fold
• 53 (28%) sit at 5% or above, 39 (21%) below 1%, and only 6 below 0.5%
• The database's strength rank predicts the dose: rank 1 → 8.07%, rank 2 → 2.69%, rank 3 → 0.76%, rho = -0.592
• The evaluation concentration predicts it better, rho = 0.679
• But the evaluation concentration falls 100-fold while the real dose falls only 5.2-fold. It is a ranking, not a dosing table

The question beginners ask most, and the one the database never answers

"How much of this do I put in?"

Every field in the database answers some other question. The CAS number answers who it is, substantivity answers how long it lasts, strength answers how loud it is. No field says "use 3%", because that depends on what you are making.

The corpus can get close, though. I have 954 public demonstration formulas holding 17,481 component records. Every one of them carries complete weights, so each material's share of each formula can be computed directly.

How I computed it

1. For each formula, sum all component amounts as the denominator
2. Compute each material's percentage share of that formula
3. Keep only materials appearing in 20 or more formulas (below that the median wobbles too much)
4. For each material, take the median of its shares across every formula it appears in

Step 3 leaves 187. Step 4 uses the median rather than the mean because the distribution is right-skewed: the same material may be dosed at 60% as the lead in a single formula, and that one entry drags the mean.

The spread is narrower than I expected

I had assumed four orders of magnitude. What I found:

Band	Materials	Share
≥ 10%	11	6%
5–10%	42	22%
1–5%	95	51%
0.5–1%	33	18%
< 0.5%	6	3%

Half of all materials land between 1% and 5%. The full range is 81-fold, not the thousand-fold I imagined.

That fact is usable on its own. If your formula contains a 0.05% or a 30%, you are standing where none of these 187 materials stands. That does not make it wrong, but it earns one pause to check.

What sits at the top

Median dose	n	Material
16.16%	33	hexylcinnamic aldehyde
15.00%	26	dextro-limonene
15.00%	49	phenylethyl alcohol
15.00%	32	dipropylene glycol (DPG, a solvent)
12.67%	28	bergamot oil replacer
12.00%	21	madrox
10.50%	44	benzyl benzoate (solvent and fixative)
10.00%	274	phenethyl alcohol (the same thing, spelled differently)

Three kinds of thing: solvents (DPG, benzyl benzoate), cheap volume florals (phenethyl alcohol, hexyl cinnamaldehyde), and reconstructions standing in for a natural oil (bergamot oil replacer). What they share is carrying volume, not supplying character.

What sits at the bottom

Median dose	n	Max	Material
0.20%	31	17.0%	ambroxan
0.25%	23	5.0%	prenyl acetate
0.30%	25	10.0%	aldehyde c-9 (nonanal)
0.30%	27	2.3%	evernyl (Veramoss)
0.37%	26	3.6%	delta-damascone
0.42%	40	17.0%	damascenone
0.50%	68	3.3%	rose oxide
0.50%	80	8.3%	ethyl vanillin

The damascone family takes three of the eight places. This group shares no chemistry with the group at the top. The one thing it shares is that the database marks almost all of them strength rank 3.

The strength rank really does predict the dose

This is the thing I most wanted to confirm this round. Of the 187, 117 could be matched back to the database and carried a strength rank:

Strength rank	n	Median dose
1 (weak)	4	8.07%
2 (medium)	89	2.69%
3 (strong)	24	0.76%

rho = -0.592. Higher rank, lower dose, and the trend is monotonic.

I have audited a lot of fields, and most of them do not survive this kind of test. The strength rank survived. A subjective grade somebody assigned by smelling in a lab predicts the amount a different set of people actually weighed out.

(Rank 1 holds only 4 materials. Do not read that 8.07% as a precise figure. What actually holds up is the contrast between rank 2 and rank 3.)

The evaluation concentration predicts better, but do not copy it

The evaluation concentration (the at N % inside odor_descriptions) does better still:

Evaluation concentration	n	Median dose
100%	75	3.88%
10%	30	0.85%
1%	11	0.75%

rho = 0.679, stronger than the strength rank.

Now the misreading. Seeing "evaluate this one at 1%", it is natural to think "so I use 1%".

The evaluation concentration drops 100-fold from 100% to 1%. The real dose drops only from 3.88% to 0.75%, a factor of 5.2.

Put differently: a material you have to dilute to 1% before you can smell its shape is not actually dosed at 1% in practice. It is dosed at about 0.75%, which is barely different from a material you only have to dilute to 10% (0.85%).

The evaluation concentration answers how much you must dilute something to perceive its shape, not how much belongs in a formula. The two are physically related, both being about thresholds, but their scales are entirely different. In 2025 Wachowiak and colleagues argued in J Neurosci for exactly this kind of scale mismatch: the concentration ranges olfactory research habitually uses do not line up with the ranges that occur in nature (PMID 40044450).

A trap in the names

In the table above, dipropylene glycol appears 32 times with a median of 15.00%, while dipg appears 445 times with a median of 9.88%.

They are the same thing. The corpus writes it under two names, and counting only one of them gives a frequency 14 times off and a dose 5 percentage points off.

I have measured the scale of this problem and hit it once already in the styralyl acetate entry. It turned up again this round, and this time it landed on the top of the leaderboard.

So how much should you use

If you take one line away:

• A material with no strength grade that you do not know → start at 2% (the overall median is 2.11%)
• The database says strength rank 3 → start at 0.5%, and make a 10% dilution before weighing
• Solvents and volume materials → 10% and above is normal, no need to feel sheepish

Those three are a starting point, not an answer. The real answer comes after you smell it.

What this doesn't establish

• These 954 are demonstration formulas, not the formulas of products that sold. Demonstration formulas tend to write amounts tidily, so they may understate the extremes found in commercial work.
• The median hides differences in role. The same material dosed as the lead and dosed as background differ a lot. The max column is there to keep that visible (damascenone peaks at 17%).
• Correlation is not causation. The strength rank predicting the dose does not mean perfumers set doses by reading strength ranks. More likely both reflect the same underlying thing: how strong the material is.
• The 117 matches rely on names. The 70 that failed to match are not a random sample. They are more likely to be trade names and reconstructions, and that class skews high-dose, so the tables probably understate the overall level slightly.
• I did not measure odour thresholds. Cometto-Muñiz and Abraham measured concentration-response functions for homologous alcohols (PMID 18950650), showing thresholds vary systematically with chain length. What I used here is the amount other people put in their formulas, which has a perfumer's judgement sitting in between.