Sixteen breads from 14,400 years ago to the ones still baked this morning — each one as the evidence has it, and again as you'd actually bake it with einkorn, rye, khorasan, T65, T80 and Lev'blé.
How to read this. Every recipe has two halves. The HISTORICAL block is what the archaeology, the period text or the unbroken tradition actually supports — including where it supports nothing and I'm inferring. The YOUR KITCHEN block is a bakeable formula in your stock, at weights that work in a domestic oven.
The tan swap bar under each pair spells out the conversion: original grain on the left, your substitute on the right, and one line on why it holds or where it breaks. Some breaks are real — barley and buckwheat have no honest substitute in your pantry, and I say so rather than fudging it.
One big thing in your favour: khorasan is not merely "like" emmer. Triticum turgidum ssp. turanicum and T. turgidum ssp. dicoccum are two subspecies of the same tetraploid species, so the grain chemistry is genuinely close and most of the ancient breads here are emmer breads. The dough is not identical, though — khorasan is higher-protein and stronger than emmer, and free-threshing where emmer is hulled. Same species predicts the genome, not the handling.
| They used | You use | How close, and what shifts |
|---|---|---|
| Emmer T. turgidum ssp. dicoccum | Khorasan, wholemeal freshly milled from your berries | The best swap here. Same species, sibling subspecies — both tetraploid AABB, both nutty and sweet. But manage two differences: khorasan is free-threshing (no parching and pounding), and it is higher-protein and stronger than emmer, sitting nearer durum in handling. Hydration: Samuel measured hand-ground emmer meal at 82% and Bats' tested Egyptian formula runs at 75%; start khorasan at 70% and add. Much of that gap is milling coarseness, not the grain. |
| Wild / domesticated einkorn T. monococcum | Einkorn, wholemeal your berries, unsifted | Exact. The one grain here you can match outright. Keep the bran in — every einkorn find in the record is wholemeal. |
| Barley Hordeum vulgare, incl. bere | No honest substitute. Buy hulled barley or beremeal. | Barley is defined by what it lacks: no functional gluten at all, so it makes paste and flatcake, never a loaf. Rye is the nearest thing you own but rye's stickiness is pentosan-driven, a completely different mouthfeel. Hulled barley berries are cheap and mill fine — worth getting, since four recipes here are barley-led. |
| Free-threshing wheat Roman siligo, medieval wastel | T65 Blanche, or T80 Bise | Very good. Siligo was milled and sieved repeatedly — that's an extraction ladder, exactly what the T numbers encode. T65 ≈ the fine sieved grades (siligo, manchet, wastel); T80 ≈ the everyday grade; T110 ≈ tourte / cheat bread. |
| Rye Secale cereale | Rye, from your berries T110–T170 for the dark grades | Exact. For rugbrød and pumpernickel you want the berries whole or cracked, not just milled — those breads are built on soaked kernels. |
| Club-rush tuber Bolboschoenus glaucus | Tigernut (chufa) flour or water-chestnut flour | Better than it sounds. Tigernut is Cyperus esculentus — same family (Cyperaceae), same wetland sedge tuber, same cell walls that stay fibrous after cooking. That fibrousness is precisely why the excavators described the original as "gritty… but a bit sweet as well." |
| Oats Avena | Buy oatmeal — medium and pinhead | Cheap, and the grade matters more than the source. Pinhead gives the coarse Highland oatcake; medium gives the thin brittle one. |
| Spelt | Khorasan (nutty route) or T80 (structure route) | Spelt is hexaploid and holds more structure than khorasan. If a recipe uses spelt for flavour, go khorasan; if it uses spelt to build the loaf, go T80. |
| Durum semola | Khorasan, milled coarse and re-sifted | Good. Durum is T. turgidum ssp. durum — again the same species as khorasan. Sift out the bran and you have a workable semola stand-in. |
| Buckwheat, teff, millet | Buy them. | No wheat substitutes. All three are non-cereals or non-wheat cereals whose whole point is what they aren't. |
| Levain / barm / zyme | Lev'blé | Where a recipe wants a stiff levain (most Roman and medieval ones do), build it at 2 parts flour : 1 part water — Roman doughs ran 52–58% and a wet starter fights that. |
Natufian, Black Desert. The oldest bread anyone has ever excavated. Arranz-Otaegui et al., PNAS 2018.
Ground wild einkorn, wild barley and possibly oat, mixed with club-rush tubers. Dehusked, ground, and properly sieved — 41% of particles under 0.3 mm, finer than most later Neolithic bread. Definitively unleavened: mean void size 0.15 mm against >1 mm in a raised loaf. Baked in the ashes of a fireplace or on a hot flat stone; there are no ovens anywhere at this date.
No ratio survives. The paper gives no cereal-to-tuber proportion — the 75:25 below is my inference, weighted toward cereal because the tuber flour alone is barely palatable. The excavators tasted the tuber flour and described it as "quite gritty and salty, but a bit sweet as well."
The salt isn't evidenced any more than it is in Ötzi's cracker below — I've kept a trace here only because Bolboschoenus glaucus grows in brackish water and both excavators independently described the tuber flour as tasting salty. Drop it for strict fidelity. Expect a dense, sandy, faintly sweet cracker with a saline edge. It is not good. It is, however, close to what it tasted like — Tobias Richter: "it is probably not to our particular tastes in the present."
Reconstructed from the Iceman's gut contents. Rollo et al. 2002; Maixner et al., Current Biology 2018.
Einkorn bran dominates the cereal residue — about 45% of the tissue in the colon sample. The particle size shows fine grinding on a saddle quern, which is why Klaus Oeggl reads it as bread rather than porridge: gruel would leave coarse whole fragments. Charcoal particles are stuck to the bran, so it was baked on a hot rock or beside a fire. Barley is in the same gut, so this was not necessarily 100% einkorn.
Oeggl's own verdict on the product: "probably hard, somewhat like a cracker, and rather tough on the teeth."
No experimental reconstruction of this bread has ever been published. Anything sold as an "Ötzi bread recipe" is invention. So is the "5,300-year-old sourdough" story — the yeast revived from his body in 2026 is Glaciozyma, a cold-adapted glacier yeast that colonised the corpse, not a Neolithic bread culture.
No salt and no leaven — neither is evidenced. If you want it edible rather than accurate, 3 g salt. If you have a fire, bake it on a stone at the edge of the coals and let a little ash touch the surface; the charcoal specks are the one detail the microscopy actually gives us.
Parkhaus Opéra lakeshore settlement, layer 13. Heiss et al., PLOS ONE 2017. Two discs, 6 cm and 10.1 cm.
Find 2285 is barley alone — a single-species galette, 6 cm across, about 2 cm thick. Find 2907 is barley plus a Triticum, and carries Apium graveolens — celery. That is the first evidence for a bread condiment anywhere in the Neolithic.
Deliberately mixed grind, and the two differ: 2285 runs roughly 29% fine flour / 35% semolina-grade / 36% coarse grist, while 2907 is noticeably finer at 43% / 24% / 15%. 2285 is about 2.1 cm thick, 2907 about 3.2 cm. Both are "clearly unleavened, and therefore flat breads (galette)" — under 5% of cavities exceed 1 mm. No intact starch granules survive, so the dough was wet before it met heat. The charring itself is read as a baking accident: these are two burnt ones.
No reconstruction of these has been published. The temperatures below are borrowed from EXARC's Iron Age tannur experiments — floor 350–500 °C, walls 100–120 °C, 10 minutes.
Eskişehir, Turkey. Found beside a doorway, a piece torn off, then carbonised and sealed under sterile soil — a deposit at the ritual abandonment of the building. Karaburun et al., PLOS ONE 2025.
Emmer, with lentil "in small amounts." Oval, 12 cm across and 2.5 cm thick. The flour was not sieved — rachis fragments are still in it. SEM shows irregular air voids of varied size, which the authors read as kneaded and possibly leavened, hedged carefully. Thermal analysis puts the bake above 150–160 °C, and the starch granules kept their structure, so it was brief.
Salt isn't evidenced but isn't excluded either; drop it for fidelity. This is the recipe on this page where your grain is closest to the original — you are baking one subspecies of Triticum turgidum in place of another.
The only ancient bread with a complete, tested, numeric protocol. Adeline Bats, J. Archaeol. Sci.: Reports 2020 — 54 loaves baked in replica conical moulds. Grain identification from Delwen Samuel's microscopy on ~70 desiccated loaves.
"Almost all examined loaves are made from emmer wheat." Samuel's key working number: emmer flour needs 82% water against 69.2% for modern bread wheat. Dehusking was done by dampening the spikelets to soften the chaff, then pounding in a mortar.
Bats' winning formula, verbatim: 1000 g black emmer wholemeal, 750 g water, 170 g einkorn sourdough starter, 15 g salt. Note that Bats' own dough runs at 75%, not Samuel's 82% — the two numbers measure different meals, and the gap is mostly milling coarseness.
Moulds of local clay tempered with dung and gravel, coated inside with fine sandy clay as a release agent, laid horizontally in an open ash-and-charcoal fire kept below 500 °C, rolled with a poker every five minutes. Bats reports 60 minutes of baking between 100 and 120 °C — that is the baking environment inside the sealed mould, not the crumb, which can't exceed ~100 °C while it's still moist. Result: fully baked, well-honeycombed crumb, clean release.
Ed Wood and Mark Lehner's 1993 replica bakery at Giza found the other essential detail: without a preheated lid the bread does not bake through. Their failure mode was over-souring — the lactobacilli took over on a long stand.
Nathan Myhrvold's suggestion, after watching the Giza reconstruction, was that the original dough was batter-like — closer to muffin batter than a kneadable dough. If you want to test that reading, push to 90% hydration and pour rather than shape.
The everyday food of most Greeks. Not a bread in any modern sense — and mostly not baked at all.
The lexical split matters: ἄλφιτα (alphita) is barley meal, ἄλευρον (aleuron) is wheat flour. Maza is made from alphita, artos from aleuron, and the two never cross.
The barley was parched or roasted before milling — necessary to release the hull, and it means the meal arrives already cooked. That is why maza needed no baking. Paulus Aegineta gives the clearest ancient statement: "The maza consisted of toasted barley flour pounded with liquor, such as water, oil, milk, oxycrate, oxymel, or honied water." That list is your formulation matrix.
Athenaeus preserves a dozen named types — physte (coarse Athenian), amolgaia (with goat's milk, Hesiod's), kardamale (with cress), thridakiskai (eaten with lettuce), achilleum (very fine barley, the luxury grade), etnitas (with pulses), daratos (unleavened). Archestratus rated Lesbian barley the finest, "whiter than snow."
(a) Maza proper — no baking
Toast the barley meal dry in a wide pan over medium-high heat, stirring, until deep brown and unmistakably nutty — 6–8 minutes. Off the heat, stir in water, oil and salt to an even thick paste. Cool, then shape into cakes by hand. Eat as is, dipped in broth, or with honey and fresh cheese. Substitute any of Paulus's liquids for the water: goat's milk for amolgaia, honey-and-vinegar for the oxymel version, watered wine.
(b) Baked barley cakes — Farrell Monaco's pharmakos version
Do not knead. Combine gently to a firm dough, rest 2 hours under a damp cloth, roll to 5 mm, cut rounds, bake 200 °C for 20–30 minutes. Serve with fresh cheese, dried figs and poppy seed.
The wheaten loaf — artos, the feast-day bread against everyday maza. This is the one Greek type with enough textual detail to formulate directly.
Athenaeus preserves Chrysippus of Tyana's catalogue of loaves, sorted by leaven, oven and enrichment: zymites (leavened), klibanites (baked in the portable bell oven), ipnites (built oven), escharites (brazier, "soft and delectable… dipped in sweet wine"), spodites (ash-baked), pyrnon (unsifted, wholemeal), dipyros (twice-baked).
One genuinely technical observation survives, from Chrysippus on the artopticius: with hard yeast, it will be white, against dissolved yeast, which gives a lighter but less white loaf. That is a real observation about stiff versus liquid levain and its effect on crumb colour, made two thousand years before anyone had a word for it.
The artolaganon line: "a little wine, pepper, and milk … oil or lard." The hapalos artos ("soft bread," Cappadocian): "with a little milk and oil and sufficient salt," dough kept "quite soft," best eaten hot.
For hapalos artos instead: 500 g T65, 340 g whole milk (68%), 40 g olive oil, 10 g salt, 100 g stiff Lev'blé. Keep it deliberately slack, bake hot, eat warm — the source insists on all three.
The eight-segment Roman loaf, from the carbonised examples in the Bakery of Modestus. Four published reconstructions disagree instructively.
The 2023 interdisciplinary study (Cardenas, Yarza, Matterne & Arranz-Otaegui) makes the argument that matters: most reconstructions use roller-milled flour, which produces a fundamentally different bread. Stone-milling and sieving are the variables — not the shaping tricks. Their formula runs at 52% hydration, well under everyone else's, mixed but not kneaded, not autolysed, not folded.
A live dispute worth knowing about. The 2023 team argue the equatorial groove comes from two stacked dough discs. Farrell Monaco, analysing all 81 carbonised loaves from Modestus's oven for the Bloomsbury Handbook of Experimental Approaches to Roman Archaeology, concludes cord really was tied round the waist. Nobody has settled it. Bake it either way.
The segments were pressed, not cut — blunt impressions radiating from a central navel. The low hydration is functional: the loaf has to hold a stamped impression and eight defined wedges without spreading.
Natural History XVIII.26, "The various kinds of leaven." Four methods, two of them seasonal.
Pliny also records the Greek convention of roughly eight ounces of leaven per modius of meal — a small-percentage levain — and, separately, that Gaulish and Spanish bakers skimmed beer barm for a lighter loaf than anywhere else.
You already have Lev'blé, so treat this as an experiment rather than a need. Method 1, over 4–5 days: 2 cups millet flour (or T80), 2 cups grapes — unwashed, the bloom is the whole point — 2 cups tepid water. Crush the grapes in cheesecloth, drain the excess, and bury the pouch in the flour-and-water slurry. Keep warm, loosely covered. Once it bubbles, at day 2–3, pull the pouch and transfer to a clean jar; feed ½ cup flour and ½ cup water. Feed again on days 3–4. By day 5 it should be vigorously active.
Maintain it stiff — 1 cup taken, 1 cup flour and ½ cup water back. That 2:1 ratio is what makes it usable in the 52–58% Roman doughs; a wet starter fights them.
De Agri Cultura, ch. 74. The whole recipe, essentially in full.
Verbatim: "Wash your hands and a bowl thoroughly. Pour meal into the bowl, add water gradually, and knead thoroughly. When it is well kneaded, roll out and bake under a crock."
Note what's absent: no leaven, no salt, no quantities. Depsticius means "well-kneaded" — the kneading is the recipe. The "crock" is the testum, an earthenware dome set over the loaf on the hearth and heaped with coals.
Knead hard for a full 10 minutes — this is the only instruction Cato gives, so honour it. Flatten to ~2 cm. Bake under a preheated Dutch oven or heavy inverted pot on the oven floor at 220 °C for 30–40 minutes. Add 8 g salt if you want it edible by modern standards, but it isn't in Cato.
Two more from the same chapters, if you're curious. Libum (ch. 75): 655 g fresh cheese to 327 g flour, one egg, baked slowly under a crock on bay leaves — a 2:1 cheese-to-flour ratio, which is why it's a cheesecake with flour in it rather than a bread. Placenta (ch. 76): the honey-and-sheep's-cheese layer cake. The EXARC replication found that the tracta pastry sheets dissolve into the filling unless you use the soaked-farro paste — that step raises hydration from 48% to 75% and is the difference between layers and mush.
The twice-baked biscuit that fed the legions. Named separately from bread in Constantius's 360 CE constitution: "buccellatum and bread, wine and vinegar."
Pliny lists panis militaris castrensis among the bread types and notes it is heavier than other kinds. Zonaras glosses buccellatum as a ring-shaped roll — the annular form matters, for stringing and for drying evenly.
The number to bake to: rebaking 850 g of bread yields about 650 g of biscuit, roughly a 24% weight loss on the second bake, with minimal nutritional loss and a keeping time of a month or more. No fat, ever — fat goes rancid, which is the whole reason it isn't there.
Rye and wheat sown together and threshed together — crop insurance that became a bread. Méteil from Latin mistilium, "mixed."
The legal definition still reserves the name for wheat and rye grown and threshed together, rye at 50% or more — not blended at the mill afterward. It fed rural France for eight centuries: Massif Central, Auvergne, Brittany, the Alps, the Ardennes, the Vosges. Froment was for towns and for people with money.
Gervase Markham, 1615, gives the practical instruction that period bakers actually needed, and it's the one modern recipes leave out: "because Rye is a little stronger grain than wheat it shall be good for you to put your water a little hotter then you did to your wheat."
To go fully faithful: mill your rye and wheat berries together in one pass rather than blending two flours. That's what the word actually means, and it's a distinction you're equipped to make.
The fullest period English bread method that survives, from The English Huswife. Everyday bread — one rung below manchet.
Three things worth stealing from Markham's text. The overnight sponge: strained sour leaven and warm water stirred into part of the flour until it is "as thick as a pancake batter," then "cover it all over with meal, and so let it lie all that night." Covering a wet sponge with dry flour instead of a cloth — it skins over and protects itself.
The salted leaven: the piece saved from the last batch is "well filled with salt, and so laid up to sour." Salting a levain to slow it between weekly bakes.
The maslin licence, verbatim: "you may bake any bread leavened or unleavened whatsoever whether it be simple corn, as Wheat or Rye of itself, or compound grain as wheat and rye, or wheat, rye and barley, or rye and barley." The whole medieval grain ladder, in one sentence, treated as one method.
Note also that the Assize of Bread ladder — simnel, wastel, cocket, whole wheat, treet, horsebread — is a sieving ladder, not a recipe ladder. The wheaten grades are the same dough. What changes is the bolting cloth, and therefore the extraction rate. Which is exactly what your T numbers are.
For manchet instead — the top of the ladder — go T65 at ~58% and keep it deliberately stiff — the 1594 text says "as hard as ye can handle it… without any more liquor," and read literally its quantities come out around 33%, which is damp crumbs rather than dough, so the period wording is doing rhetorical work. Here's the counterintuitive part: bake it at 105 °C for an hour and keep the crust pale. Period manchet was a soft-crusted white bread, not a golden one.
Oatcakes are a ratio, not a recipe: oatmeal 100 / fat 7–15 / hot water 40–50 / salt ~1. The soda in most modern versions is a 19th-century addition; the old cake is meal, water, salt and dripping. The two-stage finish is what makes them keep — set on the girdle until the edges curl, then stand the farls in a banna-rack before an open fire to dry through.
Bannock: the old one is unleavened barley, bere or oat meal, water or buttermilk, salt, cooked on a girdle or straight on a flat sandstone bannock stane. Every "traditional bere bannock" recipe online is the post-1840s chemically-raised version, and the wheat flour in all of them is there purely to compensate for bere's near-total lack of gluten.
Bere is a six-row barley landrace, possibly Neolithic, "probably Britain's oldest cereal in continuous commercial cultivation" — still grown on 5–15 hectares in Orkney and milled at Barony Mill, Birsay.
Oatcakes
Stir the melted fat and salt into the meal, add boiling water, work fast to a stiff dough. Roll on a meal-strewn board to 3 mm and cut 15 cm rounds — this batch gives you about five — quartering each into farls. Girdle or dry pan at moderate heat — you are drying, not browning — 5–8 minutes a side, then finish in a 120 °C oven for 20 minutes to dry through. Use pinhead oatmeal instead of medium for the coarse Highland version.
Pre-modern bere bannock — reconstructed by stripping the modern additions out
No flour, no soda, no cream of tartar. Bring to a soft dough, pat to 1 cm, cook on a hot dry ungreased girdle 5 minutes a side until dry-surfaced and speckled, then dry at the oven door. Dense and crumbly by design.
Flatbrød. The Norwegian ratio is stated cleanly in the literature: 6–7 dl liquid per kg of meal — 60–70% hydration — plus 1 tsp salt per kg. Barley, oat and rye, alone or mixed. The water goes on boiling, poured over the salted meal — a scald, which is what makes gluten-free meal rollable at all. As the Norwegian sources put it: "kjevlingen er største kunsten" — the rolling is the whole craft.
Breton galette. Ancestral formula: buckwheat, water, salt. Nothing else. The egg is a modern domestic adaptation. Hydration is around 200% — this is a batter — with 4% salt, which reads as wildly high until you taste it. The 24–48 hour cold rest is the real technique: it develops the elasticity, the browning and the characteristic lacy holes. Buckwheat flour goes stale in about two months and freshness is named repeatedly as the single biggest quality variable.
Flatbrød
Pour the boiling water over the salted meal, work firm, bag it and chill until cold through — overnight is fine and better. Portion ~150 g per leaf. Dust the board heavily with barley flour and roll from the centre outward, as thin as you physically can. Bake on a dry pan or griddle at moderate heat until the surface dries, flip, finish to light brown spots. Dry at 100 °C with the oven door ajar. Stores for months.
Galette de blé noir — the crêpier's formula, no egg
Day 1: mix the flour, salt and 500 ml water. Work it by hand with a claw grip, then slap the batter with your palm for two minutes to aerate it. Cover, refrigerate. Rest 24 hours minimum, 48 better. Day 2: out an hour ahead, whisk in the remaining 500 ml. Cook on a very hot greased pan or billig, ~200–240 °C, about a minute, traditionally one side only.
The bread that closes the loop. Einkorn survives as a living crop in Kastamonu — over 900 farmers still grow siyez.
Be clear about what's traditional here. The Slow Food Presidium in Kastamonu is for siyez bulgur, not bread: spikelets boiled 20 minutes, cooled, sun-dried, cracked at the mill until "split in two like the wings of a mosquito," re-dried, eaten as pilaf. That's where the calories went. A Turkish food-science paper on siyez in breadmaking says flatly that its "bread production quality was found to be low," and the marketed 100% siyez loaf is a modern revival product.
Where einkorn bread genuinely was traditional and genuinely was 100%, the record points to the western Rif in Morocco — einkorn as skaliah, "used for bread, either pure or mixed with other cereal and legume flours" — and to living memory in Transylvania and Slovakia, "valued for its specific taste." In all of them it's flat or griddle-baked, not a risen loaf.
The more historically honest version: take the same dough, divide into 6, roll to 5 mm and bake on a stone at 280 °C for 3 minutes a side. That flat, dense, deep-yellow thing is what 100% einkorn bread actually was for most of its history — and it's what the Moroccan, Transylvanian and Anatolian traditions all point at.
Worth knowing, because recipes for all three circulate:
Baking is the hobby; the software is back at Luke Pezet. Corrections welcome — lpezet@gmail.com.