Bakery shelves and flour bags carry a growing vocabulary of age: heritage wheat, heirloom wheat, ancient grain, landrace. Only one of those terms has a body of scientific literature attempting to pin it down, and even there the definition is contested and moving. Knowing which is which is the difference between an informed purchase and a story.
Landrace has a definition, and it has been rewritten
The word entered agricultural science with von Rümker in 1908, and for many years afterwards it meant something quite specific: cultivars that had evolved without conscious selection. That framing puts landraces outside human intention, shaped only by environment and chance.
Later definitions loosened it. In 1998, Zeven described an autochthonous landrace as one grown for a long period in the farming system concerned, and made a point that is easy to miss: such a population is continuously contaminated by genotypes from a few other landraces or cultivars, and adapts through that contamination. An allochthonous landrace, one recently introduced from elsewhere, he expected to be rare.
In 2005, Villa and colleagues proposed what became the most cited formulation: a dynamic population of a cultivated plant that has historical origin, distinct identity and lacks formal crop improvement, as well as often being genetically diverse, locally adapted and associated with traditional farming systems. That definition was subsequently adopted in a 2007 report by a task force on farm conservation and management.
A 2017 review in Frontiers in Plant Science then argued that two phrases in it, lacking formal breeding and traditional farming systems, are ambiguous. Their replacement definition describes cultivated varieties that have evolved and may continue evolving, using conventional or modern breeding techniques, in traditional or new agricultural environments within a defined ecogeographical area and under the influence of local human culture, with selection by farmers or breeders, conscious or unconscious.
That is a substantial shift. Under the older reading, using a modern breeding technique disqualified a population from being a landrace. Under the revised one it does not, provided the population remains locally embedded and continues to change.
Heritage and heirloom do not
Searching for the equivalent authority behind heritage variety or heirloom variety comes up empty. There is no widely accepted scientific definition distinguishing either term from a landrace, or from an old commercial cultivar, or from a variety released eighty years ago by a public breeding programme. In practice, the terms mark a claim about age and about not being modern, without specifying either.
The same 2017 review names the consequence directly. It observes a commercially promoted belief that landraces generally possess superior nutritional and sensory qualities, a belief the authors characterise as a flavour of the past, and notes that this belief has begun to influence the concept of landrace itself. When a category acquires market value, pressure follows to widen it.
None of this means old wheats are uninteresting. It means the label alone tells you nothing verifiable, and that a specific variety name plus a source is worth more than any adjective.
What is actually conserved, and how well documented it is
Where the vocabulary is vague, the collections are countable. Genesys, the portal aggregating plant genetic resources data from contributing genebanks, reports over 4.5 million accession records in total.
For wheat specifically, its crop page states that almost 320,000 wheat accessions are visible on the portal, of which nearly 23 per cent are landraces and about 7 per cent wild relatives. Broken down by species, the largest holdings are Triticum aestivum at 358,539, Triticum turgidum at 82,177, Triticum durum at 12,633 and Triticum monococcum at 4,461, with 11,261 Aegilops accessions among the wild relatives. By holding institution, the Mexican collection leads at 162,121, followed by the United States at 61,995, Australia at 50,426, Lebanon at 44,596 and Russia at 35,310. The page also states that over 560,000 accessions are held across 40 different collections according to the global conservation strategy for wheat, rye and triticale.
Two cautions belong with those numbers. First, the page is internally inconsistent: its text gives almost 320,000 wheat accessions while its own structured field reports 496,509. Second, an accession is a conserved seed sample, not a distinct variety. Duplicates across collections are common, which is precisely why a global strategy figure and a portal figure differ.
The data quality problem, stated by the genebanks themselves
The most useful thing published on this topic recently is not a count but an admission. In July 2026, Genesys described building an animation of 125 years of collecting compressed into 38 seconds, and set out what it had to discard to do so.
The portal holds more than 970,000 landrace accession records. Removing records with no collection year, or with an impossible one, left 346,000. Only around 440,000 of the original set carry latitude and longitude coordinates. Records satisfying both conditions, a valid year and a location, numbered 228,000.
So less than a quarter of landrace records in the world's largest aggregated genebank database can be placed in both time and space. That is not a criticism of the genebanks, which have inherited a century of inconsistent passport data. It is a direct constraint on any claim that a particular old wheat comes from a particular place at a particular time.
What a national collection actually offers
The institutional side is more concrete. The United States National Small Grains Collection, part of the Agricultural Research Service, maintains worldwide diversity in wheat, barley, oat, rice, rye, triticale and wild relatives including Aegilops.
Its history is worth knowing because it explains what such collections are for. Small grains germplasm collection began around 1897 with the establishment of the Seed and Plant Introduction Office. Cereal investigators in the Bureau of Plant Industry maintained the material through the first half of the twentieth century. The collection was formally organised in 1948 at Beltsville, Maryland, and moved to its present site at Aberdeen, Idaho, in October 1988. Passport and evaluation data are maintained in the Germplasm Resources Information Network, and seed can be requested online.
The collection is explicit that seed available for distribution is a limited resource, and that repeat requests for the same accession will not be honoured. These are research collections supporting breeding and study, not a supply chain for artisan flour.
How to make a defensible claim about old wheat
Three questions do most of the work.
Which variety, by name. A named variety with a documented origin can be checked against genebank records. Heritage wheat cannot.
What is the evidence for the property being claimed. If a flour is sold on flavour, nutrition, or digestibility, ask what was measured and in what comparison. The 2017 review's warning about a flavour of the past applies precisely here, and the answer is often that nothing was measured.
Is the population still evolving. Under the revised definition, that is the interesting question about a landrace: whether it is a living, locally adapted population or a genebank sample grown out. Both are legitimate, but they are not the same thing, and only one of them is what the word landrace was coined to describe.
