DVF: how to find recent French property sale prices, free

A guide to the dataset that tells you what every property in France actually sold for, with practical instructions for using it directly and an honest take on its limits.

Updated May 2026. No agent ever pays us anything.

If you arrived here looking for the French equivalent of the UK Land Registry sold-price data, you’ve found it. DVF is open, free, comprehensive, and almost no buyer researching French property knows it exists. This page explains what it is, how to use it, and why a raw median on the data lies.

The short answer

DVF (Demande de Valeurs Foncières) is the French government’s mandatory register of every property transaction recorded by a notaire since 2014. It’s free, public, and contains roughly 3 to 5 million new transactions added each year, building on the cumulative dataset since 2014. You can browse it on the official Etalab map at app.dvf.etalab.gouv.fr, or download the raw CSV files from data.gouv.fr. For most buyers, the map is enough. For comparable-sales analysis, you need either patience with French CSVs or a tool that does the work for you.

This page explains what DVF is, what it includes (and doesn’t), how to read it, and how to use it well as a buyer.

What DVF actually is

DVF stands for Demande de Valeurs Foncières, literally “request for land values.” It’s the open-data publication of every property sale that passes through a French notaire, which means every legitimate property sale in France. The dataset is maintained by the Direction Générale des Finances Publiques (DGFiP) and published openly through Etalab, the French government’s open-data office.

Coverage:

Geographic: all of metropolitan France, plus DOM-TOM (overseas departments) where the notaire system applies. Alsace-Moselle has slightly different rules and partial coverage.
Temporal: all transactions from January 2014 forward.
Volume: several million new transactions added per year on top of the cumulative record since 2014. The total dataset is large; the precise count depends on the release date.
Refresh: twice a year, in April and October. The April release covers transactions through the end of the previous calendar year; October covers through end of June.

What’s in each record:

Sale date.
Sale price.
Property type (Maison, Appartement, Local, Terrain, etc.).
Surface (surface réelle bâtie and surface terrain).
Number of rooms (nombre de pièces principales).
Address (street and commune).
Geocoded coordinates (in the DVF géolocalisée version).
Cadastral parcel reference.

What’s not in each record:

The condition of the property at sale. A handsomely renovated house and a dilapidated equivalent show up identically.
The energy rating (DPE) at sale.
Whether the property had a pool, a view, original stone walls, or any other feature that materially affects price.
The buyer’s or seller’s identity.
Whether the sale was at arm’s length or among family members.
Whether the sale was a viager (life tenancy) or otherwise unusual.

Those gaps matter. A raw price-per-square-metre median across DVF records lies in predictable ways, which we’ll get to.

Why DVF exists (and why it didn’t, until 2019)

For most of recent French history, property sale prices were technically public (the cadastre was queryable) but practically inaccessible. You could request a paper extract from a notaire or centre des impôts, but there was no online database. Buyers operated on listing prices and agent estimates, both of which lie in known directions.

The 2018 loi pour un État au service d’une société de confiance (ESSOC law) committed the government to open-data publication of property sales, and the first DVF release landed in April 2019, covering transactions back to 2014. This was a structural shift: France went from one of the more opaque property markets in Europe to one of the more transparent, in a single regulatory move.

The follow-on shift was the geocoded version, DVF géolocalisée, which adds latitude and longitude to each record. This is what makes radius-based comparable-sales analysis possible. The non-geocoded version is harder to use but identical in content.

How to use DVF directly

Option 1: the Etalab map

The simplest way to use DVF is the official Etalab map at app.dvf.etalab.gouv.fr.

The flow:

Search by address or zoom into the area you care about.
Click any cadastral parcel.
See a list of every transaction on that parcel since 2014, with date, price, surface, and property type.
Compare against neighbouring parcels by clicking around.

This works well for spot checks. “What did the house next door sell for?” takes about thirty seconds. “Has this specific property changed hands recently and at what price?” is just as fast.

It works less well for comparable-sales analysis. The map doesn’t filter by surface, doesn’t adjust for condition or energy rating, doesn’t aggregate, and doesn’t show you whether you have enough comparables to draw a conclusion.

Option 2: the raw CSV files

For analytical work, you can download the CSV files directly from data.gouv.fr. They’re released by year, with separate files per geographic region. Each file is large (sometimes hundreds of megabytes). The columns are well-documented but in French.

This option requires:

Comfort with CSVs and a tool that handles them at scale (Excel won’t load the larger files; a database, Python, or R is more appropriate).
Reading French column names.
Doing the geographic filtering, surface filtering, type filtering, and adjustment work yourself.

It’s the right path if you’re doing systematic research across many properties. It’s the wrong path for a buyer evaluating one or two specific listings. Most buyers do not have a comfortable Saturday afternoon’s worth of CSV manipulation in them after also reading guides about compromis de vente and notaire fees.

Option 3: a tool that wraps DVF

This is what Adresse.ai does. We ingest the DVF dataset, layer the ADEME energy-rating data on top, apply regional adjustment factors for the things DVF doesn’t capture (condition, pool, stone character, historic centre, view), filter intelligently by property type and surface, and produce a defensible price range with the underlying transactions visible. Full methodology at /how-it-works.

We’re not the only DVF-wrapping tool that exists. MeilleursAgents and SeLoger Estimate both blend DVF with their own listing-price models, but the methodology is opaque and the framing is seller-side. Patrim is the official government tool that uses DVF directly, but requires a French tax number to log in. PriceHubble uses DVF in its institutional AVM but isn’t available to consumers. The honest grades on each are at /guides/zestimate-for-france.

Why a raw DVF median lies

The single most important thing to understand about using DVF as a buyer: a price-per-square-metre median across the dataset gives you a misleading number.

The reason is structural. DVF doesn’t carry property condition, energy rating, pool presence, stone character, historic-centre location, or view. So when you compute a median €/m² for “houses in this commune,” you’re averaging across:

A handsomely renovated DPE-A property with a pool and a view.
A “to refresh” DPE-D property with no pool and no view.
A passoire thermique (G-rated) property facing major renovation costs.

The unadjusted median tells you approximately what the typical property in that commune sold for. It does not tell you what the specific property you’re evaluating should sell for.

A worked illustration. Consider two sales in the same village:

House A: 200m², good condition, DPE C, pool, view. Sold €820,000.
House B: 200m², “to refresh,” DPE F, no pool, no view. Sold €510,000.

Raw average €/m² across these two: €3,325. But if you’re evaluating a third property that’s 200m², good condition, DPE D, with a pool but no view, the raw average tells you almost nothing useful. You need to adjust each comparable toward your target’s profile before you average. This is the work the comp-engine does, with adjustment weights tuned per region.

When you look at unadjusted DVF medians (the kind shown on real estate news sites), keep this in mind: the number is a background figure, not a per-property estimate.

What DVF doesn’t tell you, and how we compensate

Five things DVF can’t see, and how a serious comparable-sales analysis handles each:

What DVF lacks	How a real analysis handles it
Property condition	Language cues from the listing description, plus the DPE record, used to estimate condition relative to the average comp
Energy rating (DPE)	Joined from ADEME’s open data register on the address
Pool presence	Inferred from listing text plus aerial imagery for the comp pool average
Stone character / cachet	Tuned regionally; explicit listing language is the strongest signal
Historic-centre location	Computed from the parcel position relative to local secteur sauvegardé boundaries

Adresse.ai does each of these. A spreadsheet from raw DVF doesn’t, unless you do them all by hand.

The DVF lag

DVF lags real-time by approximately six months. The mechanism: a notaire records a transaction at signing, the record propagates through the fichier immobilier to the DGFiP, and the DGFiP releases the next semi-annual DVF update. The April 2026 release contains transactions through approximately end of October 2025. The October 2026 release will contain transactions through approximately end of April 2026.

For trend-level analysis (is the market going up or down?), this lag is fine. For knowing whether last week’s sale next door was at €620k or €640k, the lag is a real limit, and no tool eliminates it. The next-best signals during the lag period are Notaires de France monthly indices and listing-portal price-cut histories, both of which operate on shorter cadences but are noisier.

Will DVF coverage improve?

The dataset is one-way. Each release adds new transactions; nothing gets removed. The cadence is unlikely to accelerate (the lag is structural, not bureaucratic), but the data quality has gradually improved since 2019, with more parcels geocoded and fewer obvious data-entry errors.

The natural next-generation upgrade would be live transaction publication on a multi-day lag rather than semi-annual batch. The technical infrastructure to do this exists; the regulatory and political appetite to push it through is the bottleneck. Don’t hold your breath, but don’t rule it out by 2030.

What this means for you

If you’re using DVF directly as a buyer:

Use the Etalab map for spot checks. Click the parcel, see what it sold for. Useful and free.
Don’t trust raw price-per-square-metre medians. They include the run-down properties, the renovation projects, and the prime examples all together.
Account for the lag. A property that sold last summer is probably in the dataset by now; a property that sold last month isn’t.
For comparable-sales analysis on a specific listing, use a tool that adjusts for the things DVF doesn’t capture, or do the work by hand.

If you’re earlier in your search and just want to understand what’s been happening in a region:

The Notaires de France market reports give you the headline trends quarterly.
INSEE publishes commune-level long-run series.
The IGEDD price-of-housing index gives you the long-run national and regional benchmark.

For a specific property, run a free estimate. The Adresse.ai output shows you the comp pool transparently, with each sale linked back to the public DVF record.

Questions

Is DVF the same as the cadastre?

No. The cadastre is the parcel registry: who owns what, where the parcel boundaries are, what type of land use is registered. DVF is the transaction register: who paid what, when. Both are open data; they answer different questions.

Can I see who bought a house?

No. DVF anonymises buyer and seller. The cadastre shows current ownership but is harder to query at scale.

Does DVF cover commercial property?

DVF includes Local and Terrain property types alongside Maison and Appartement. So commercial transactions are included, but the dataset is most useful for residential.

Why do some prices in DVF look wrong?

Three common reasons: the sale was a viager (life tenancy with reduced upfront price), an intra-family transfer at below-market price, or a sale where the recorded price excluded shared-ownership components like business assets. Adresse.ai filters obvious noise (€1 transfers, viager sales, bulk transfers); doing the same on raw DVF takes attention.

How is the geocoded version different?

DVF géolocalisée is the same dataset with latitude and longitude added per record. Etalab maintains both versions. The geocoded version makes radius-based comparable-sales analysis trivial; the non-geocoded version requires you to join on commune and street.

Is DVF used by professional valuers?

Yes, extensively. Notaires use it as a primary input to their own avis de valeur. Banks use it through PriceHubble and similar AVMs. Estate agents reference it. The data has gone from “open but unfriendly” in 2019 to “the spine of French residential valuation” in 2026.

Try it on your listing

Adresse.ai uses DVF as the primary data source for every estimate. The first one is free.

Run an estimate →