Findings
Europe documents its private heritage unevenly. This is a measurement of by how much.
A pan-European corpus of historic buildings estimated to be privately owned, built on one method across 45 countries, with its error rates measured against a national register and published beside it.
The gap this answers
Europe's privately owned historic houses have political representation and almost no data. The European Historic Houses federation speaks for 27 national associations, and its headline figure of more than 50,000 privately owned historic houses is a membership extrapolation with no per-building record behind it. The European Commission's own commissioned study says as much itself.
“…the lack of comprehensive, systemic data on this sector.”
“…not sufficiently robust to allow for any cross-country comparison.”
Heritage Houses for Europe: Exchange & Innovate (2019), commissioned by the European Commission and co-authored by European Historic Houses, the European Landowners' Organisation and IDEA Consult. The first quotation states what the study exists to address; the second, from §3.4, describes the national data available to it.
This page is that cross-country comparison, made on one method.
What the corpus shows
19.0×
spread per 1,000 km² of land
Highest BE at 86.6; lowest TR at 0.1.
7.5×
spread per million inhabitants
Highest LV at 277.6; lowest TR at 1.2.
Ratio of the 90th to the 10th percentile, across the 36 countries holding 50 or more documented sites. Measured August 29, 2026.
The corpus
- Documented sites
- 83,382
- Countries
- 45
- With an ownership estimate
- 65,927
- Estimated privately owned
- 23,466
Read the coverage column first
The share of a country's sites carrying any ownership estimate ranges from 27% to 94%. That variation is not random — it follows where the curation pass has reached.
So a low count of estimated private houses can mean that a country holds few, or that we have not yet classified what it holds. Those two are not distinguishable from this table, and the estimated-private figures should never be read without the coverage beside them.
By country
Ordered by documented sites. The bar behind each count reads its density per 1,000 km², so the spread above can be checked against the distribution rather than taken on trust.
| Country | Sites | Coverage | Est. private | Per 1,000 km² | Per M people |
|---|---|---|---|---|---|
| FRFrance | 17,879 | 86% | 7,839 | 33.2 | 260.2 |
| DEGermany | 12,684 | 67% | 2,799 | 36.3 | 151.9 |
| ITItaly | 9,550 | 83% | 2,720 | 32.3 | 162.1 |
| GBUnited Kingdom | 7,684 | 82% | 2,311 | 31.8 | 110.6 |
| ESSpain | 7,228 | 81% | 1,527 | 14.5 | 146.4 |
| PLPoland | 4,182 | 90% | 952 | 13.7 | 114.8 |
| BEBelgium | 2,641 | 84% | 895 | 86.6 | 221.2 |
| CZCzechia | 2,421 | 88% | 378 | 31.4 | 222.4 |
| ATAustria | 2,239 | 73% | 760 | 27.1 | 243.2 |
| GRGreece | 1,832 | 43% | 28 | 14.2 | 175.9 |
| CHSwitzerland | 1,665 | 65% | 308 | 42.1 | 183.1 |
| NLNetherlands | 1,557 | 75% | 331 | 46.2 | 86.1 |
| PTPortugal | 1,302 | 70% | 356 | 14.2 | 120.5 |
| IEIreland | 1,249 | 72% | 296 | 18.1 | 227.7 |
| SESweden | 1,056 | 94% | 381 | 2.6 | 99.7 |
| HUHungary | 1,039 | 92% | 226 | 11.4 | 109.2 |
| DKDenmark | 1,017 | 96% | 518 | 25.4 | 169.2 |
| SKSlovakia | 826 | 93% | 130 | 17.2 | 152.6 |
| UAUkraine | 666 | 69% | 26 | 1.1 | 17.1 |
| RORomania | 565 | 67% | 108 | 2.5 | 29.7 |
| LVLatvia | 513 | 95% | 87 | 8.2 | 277.6 |
| LTLithuania | 501 | 95% | 95 | 8.0 | 173.4 |
| FIFinland | 448 | 93% | 149 | 1.5 | 79.3 |
| HRCroatia | 397 | 76% | 56 | 7.1 | 102.4 |
| SISlovenia | 348 | 89% | 43 | 17.3 | 163.3 |
| RSSerbia | 296 | 23% | 17 | 3.5 | 45.2 |
| BGBulgaria | 278 | 17% | 5 | 2.6 | 43.2 |
| EEEstonia | 278 | 94% | 56 | 6.5 | 203.4 |
| BABosnia & Herzegovina | 160 | 31% | 2 | 3.1 | 51.0 |
| ALAlbania | 137 | 80% | 2 | 5.0 | 58.3 |
| MKNorth Macedonia | 137 | 1% | 0 | 5.4 | 75.2 |
| NONorway | 121 | 66% | 20 | 0.3 | 21.6 |
| TRTürkiye | 99 | 67% | 4 | 0.1 | 1.2 |
| MEMontenegro | 88 | 14% | 3 | 6.5 | 141.2 |
| LULuxembourg | 73 | 86% | 22 | 28.4 | 106.3 |
| MDMoldova | 72 | 51% | 5 | 2.2 | 30.5 |
| CYCyprus | 48 | 31% | 0 | 5.2 | 35.0 |
| XKKosovo | 45 | 64% | 0 | — | 28.5 |
| MTMalta | 37 | 68% | 9 | 115.6 | 63.8 |
| LILiechtenstein | 6 | 50% | 1 | 37.5 | 146.3 |
| ISIceland | 5 | 20% | 0 | 0.0 | 12.7 |
| ADAndorra | 4 | 75% | 0 | 8.5 | 48.2 |
| SMSan Marino | 4 | 75% | 0 | 66.7 | 117.3 |
| VAVatican City | 4 | 100% | 0 | — | — |
| MCMonaco | 1 | 100% | 1 | 479.8 | 26.1 |
† Fewer than 50 documented sites, and so excluded from the percentile statistics above, where a single building would move the figure visibly. Every country stays in the table.
Method and error rates
Every site is drawn from OpenStreetMap and Wikidata, merged on one pipeline, de-duplicated by location and name, and given an ownership estimate by a published rubric. The estimate is an estimate: it is not a title search, and the corpus says so wherever it is published.
Accuracy was measured against the French national heritage register, against thresholds registered before the numbers existed: precision 0.861 and recall 0.668 on the strict, name-agreed reading. The shortfall is abstention rather than error — 897 sites declined against 90 answered wrongly — and where the classifier does commit, recall reaches 0.957.
Download the corpus
The whole corpus is published as one versioned bundle — CSV, GeoJSON and newline-delimited JSON, with a data-package descriptor and checksums. The query interface answers filtered subsets; the bundle is how you take all of it.
Released under the Open Database License (ODbL), the licence its OpenStreetMap sources require.
How to cite
Šostok, J. (2026). herimap — European heritage ownership corpus (v1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22149838
What is not measured yet
The corpus is the first of four planned layers, and the only one with data in it. The others are named here so that nothing on this page is mistaken for something it is not.
- Layer 2 · Operational signalEvent kinds, offering mix, seasonality and occupancy. Modelled in the platform, with no rows recorded.
- Layer 3 · Survey evidenceTwo instruments are drafted and no responses have been collected. No survey infrastructure exists in the platform yet.
- Layer 4 · Visitor provenanceWhere visitors travel from. Absent and not modelled; it carries the highest disclosure risk of the four, and its design starts from that.
Measurement of layers 2 to 4 begins at public launch. Until then, no claim on this site rests on them.