Domesday Book counted mills place by place. Nine hundred and forty years later, are the traces of mills on today's maps in the same places? And when the Industrial Revolution built its textile mills, did it reuse the old mill sites or go somewhere new?
Tealby is a village in Lincolnshire. In 1086 Domesday Book recorded 52 households there and 16.5 mills, more than almost anywhere else in England. Open a modern map and there is still a mapped watermill 667 m from the centre of its Domesday grid square. Is that a coincidence, or does a good mill site, once found, stay a mill site for nine centuries?
Mills are a gift to anyone interested in how long places last. A watermill needs a particular spot: enough fall in the stream, a place for a weir and a leat, a crossing, a lord willing to invest. Once the dam and the channel were dug, rebuilding on the same spot was cheaper than starting again. Domesday Book recorded the mills of 1086 manor by manor, so it gives a national "before" picture, and modern open data give a rough "after".
So I counted every mill in the Domesday entries, placed them, and set them against three modern traces of mills: buildings mapped as watermills, windmills and other mills in OpenStreetMap and Wikidata, and features whose names contain the word Mill on the Ordnance Survey's gazetteer. Then I asked a second question, about the Industrial Revolution: did the textile mills of the 1700s and 1800s go to the old mill places? The short answer is that the watermills mostly did, and the textile mills mostly did not.
Each brown disc is a place with a mill in 1086, sized by the number of mills its entries record. A dark ring means a mill building is mapped within 1 km today. Switch on the modern layers to see the traces themselves. Tap a place, or search for one, to see its Domesday figures and how far it is to the nearest trace of a mill.
The figure most often quoted, for example in Wikipedia's article on Domesday Book, is 5,624 mills, and Wikipedia's article on the watermill adds that later research puts it at 6,082, a minimum because the north was never properly surveyed. Adding up the mills in the Hull Domesday Project's coding of every entry, I get 6,172.25, in 4,160 entries: 5,164.25 in Great Domesday and 1,008 in Little Domesday (Essex, Norfolk and Suffolk).
That is 548 more than the traditional figure and 90 more than the later one. I cannot reconcile the totals line by line, but the reasons are easy to see. Domesday often records a share of a mill: 241 entries have half a mill or another fraction, and different counters treat these shares differently. The same mill can appear in two entries when two lords shared it. And Yorkshire, where Domesday is thin, contributes 140. So my count is a count of what the entries say, not a census of millwheels.
Norfolk, Lincolnshire and Wiltshire lead on raw numbers, but Wiltshire stands out once you allow for how many people there were: 5.3 mills for every 100 recorded households, against 2.1 in Norfolk. That is the highest rate of any county. To put the 6,172 mills on a map I give each entry's mills to the places it describes, which leaves 3,390 places with at least part of a mill, out of 11,892 Domesday places that have an entry and a location (places of the same name in the same grid square are merged).
Now the test. For the 11,892 Domesday places that an entry describes, I measured the distance to the nearest modern trace of a mill. The answer depends on the kind of trace, and that turns out to be the most telling part.
Within 1 km, 4.8% of Domesday mill places have a mapped watermill today, against 1.2% of places without a Domesday mill: 3.9 times as many. Mill places are bigger, though, and some counties are mapped far better than others, so a fairer yardstick is the share among places without a mill in the same county and of similar size (the same fifth by 1086 households). On that local expectation the mill places have 4.7 times the expected rate. A regression that also allows for distance to a watercourse and the steepness of the ground gives an odds ratio of 3.59 (95% interval 2.62 to 4.92).
The windmills are the control. They are not where the Domesday mills were, and they should not be.
Windmills make a natural test of whether this is just "big villages have more of everything". Domesday's mills were driven by water and, according to Wikipedia's article on the windmill, the earliest certain reference to a windmill in northern Europe dates from 1185, a century after the survey. If old mill places only had more mills because they were bigger or better mapped, they should have more windmills too. They do not: 2.6% have a mapped windmill within 1 km against 2.4% for other places, and after allowing for county and size the odds ratio is 0.75 (0.57 to 1.00). The persistence belongs to water power, not to mills in general.
The pattern rises steadily with the number of mills. Places whose entries record three or more mills have a mapped watermill within 1 km 10% of the time, 8.4 times the rate for places with none. Counting any mapped mill building (water, wind, textile or unspecified) the share is 16% against 7.4% (odds ratio 1.95), and for features named Mill on the Ordnance Survey gazetteer it is 5.0% against 3.8% (odds ratio 1.69).
Part of the reason is plain geography. The median Domesday mill place sits 399 m from a watercourse on the Ordnance Survey's river network, against 548 m for places without a mill, and 60% are within 500 m of one, against 46%. But the regression already allows for that, and the watermill effect remains.
Among the 25 biggest, 14 have a mapped mill or a Mill name within 1 km. Derby, top of the list with 25 mills, has no mill building mapped within 1 km of its grid square in these data, and nor do Dorchester, Sherborne or Chippenham. Towns are where the open data are least likely to tag an old mill as a mill.
At county level the link is much weaker. The rank correlation between Domesday mills and mapped mill buildings today, county by county, is 0.30 (p = 0.080, 34 counties), and for watermills alone 0.25. Domesday mills track the number of recorded households very closely (0.82), but modern traces track how keenly each county's mills have been mapped and listed.
The outlier at the top is Yorkshire, with 430 mapped mills against 140 Domesday mills, and the dot high on the left is Cheshire, whose nearest Domesday places pick up the mill towns of south Lancashire (78 textile mills). That is a useful warning. The place-level result works because it compares places within the same county, where mapping habits are shared. Counting mills across counties mostly measures the mappers.
The second question is about the great textile mills. Did the cotton, wool, silk and flax mills of the 1700s and 1800s take over old corn-mill places, or go somewhere new? Wikidata gives me 113 textile mills in England with a location, named by their class or their label. They are overwhelmingly northern: 84 are in the North West and 83% in the North West or Yorkshire.
The median watermill on today's map is 1.6 km from a Domesday mill place, and 57% are within 2 km. The median textile mill is 20.4 km away, and only 6% are within 2 km. The handful that are close are mostly in the south and west: a flax mill at South Petherton, silk mills at Tring, Chipping Campden and Whitchurch, Pinsley Mill at Leominster.
The main reason is not that mill-builders shunned old sites. It is that Domesday recorded almost no mills where the cotton industry grew. In my Domesday data the North West has 337 places and just 15 with a mill; Yorkshire and the Humber has 134 mill places out of 1,757, while the South West has 782. Even comparing like with like inside the North West and Yorkshire, the median textile mill is 22.4 km from a Domesday mill place against 8.1 km for a watermill there, and none of the 94 northern textile mills is within 2 km of one.
The setting tells the same story. Watermills hug their streams: the median one is 19 m from a watercourse on the OS network and 89% are within 100 m. Only 35% of the textile mills are. Many of the textile mills in Wikidata are big town mills, such as Houldsworth Mill in Reddish, Kingston Mill in Stockport and Cavendish Mill in Ashton-under-Lyne. Valley mills like Cressbrook in Derbyshire are in there too, but they are a minority of what is mapped, and Wikidata's list is far from complete. So the honest answer is: the textile mills grew mostly in a different part of England, often away from the riverbank, and the traces of mills that cluster around the old Domesday mill places are watermills, not factories.
These are traces, not a census. OpenStreetMap and Wikidata record the mills that someone has mapped or listed, and coverage is uneven: some counties are mapped far more fully than others. The comparisons are therefore made within counties, and the county-level result shows how much mapping matters. A place with no trace may well have had a mill until recently.
Places are approximate. Domesday places sit at the centre of their 1 km grid square, and a Domesday "place" can be a whole parish's worth of manors, so distances under about 700 m are within the noise. A mill could stand 2 km from its village and still belong to it.
Domesday is uneven too. Yorkshire's record is thin, the far north was not surveyed, and Lancashire appears only in part. Mills shared between manors appear as fractions, and the allocation of an entry's mills to the places it describes is a rule of thumb.
Textile mills are a small, biased sample. I could identify 113 textile mills in England from Wikidata classes and labels; many more mills in the data are unspecified, and I cannot tell water-powered from steam-powered mills. The textile result is about where mapped textile mills stand, not a full history of the industry.
Same place, not same mill. A modern watermill near a Domesday mill place is consistent with continuity of the site, but it does not prove that the building stands on the 1086 site. Many mills were rebuilt or moved along the same stream.
Domesday. I summed the mills field for every entry in the realm-domesday repository's reference files, which carry the Hull Domesday Project's coding of Great Domesday and figures parsed from the translation of Little Domesday. Each entry's mills are shared equally among the places it describes (the rule used in my Domesday Destiny study). 98.5 mills sit in entries that describe no mapped place.
Modern traces. OpenStreetMap objects tagged as watermills, windmills, tide mills or mills (1,101), Wikidata items in the United Kingdom of a mill class with coordinates (3,331), and OS Open Names features with Mill or Mills as a word (1,617). Kinds come from tags, classes and labels: water (watermill, tide, corn and flour mills), wind (windmill, post, smock and tower mills), textile (cotton mill class, or a label naming cotton, wool, worsted, silk, flax, linen, lace, spinning or weaving), other industry, and unspecified.
Tests. For each Domesday place with an entry I measured the straight-line distance to the nearest trace of each kind. Local expectation: the share among places without a mill in the same Domesday county and the same fifth of places by 1086 households. Regressions: logistic, with county fixed effects, log households, log distance to the nearest watercourse (OS Open Rivers) and log local relief (terrain within 1 km). Textile mills were compared with other mapped mills by distance to the nearest Domesday mill place and by setting.