Methodology

This site is built exclusively on open data published by Dubai government bodies — primarily the Dubai Land Department. No listing portals, no scraping, no proprietary feeds. The full registry of datasets is on Sources.

• Registered transactions, not asking prices. Every figure reflects what was registered with the Dubai Land Department, not what is advertised on portals.
• A source on every number. Each published figure carries its dataset, sample size and freshness date. Figures that fail quality or sample-size rules are suppressed, never approximated.
• Medians, never averages. Averages are not published anywhere on this site (why).
• No investment advice. The numbers describe recorded market activity. They are not predictions, valuations or recommendations.

Data arrives through two official Dubai Land Department channels: the bulk open-data archive on Dubai Pulse (full history, 2004 onwards) and the DLD open-data gateway, which serves recent transactions. Together they form a continuous, deduplicated record.

• Weekly updates. An automated import runs every Monday: new transactions and the latest price index are loaded, every month touched by the import is recomputed from the row-level data (the full history is rebuilt after any backfill or repair), and pages are refreshed. Each page shows the date its data was last updated.
• Deduplication on official identifiers. Records are keyed on the DLD’s own transaction numbers, so the same transaction is never counted twice even when both channels deliver it.
• Audited, reversible imports. Every import run is logged with row counts and rejection stats. Superseded records are marked stale and excluded from all statistics — never silently deleted.
• The data edge anchors all windows. Trailing windows (“last 12 months”) are measured back from the most recent registered transaction in the dataset, not from today’s date — so a delayed source can never silently shrink a window.

The DLD transactions dataset records every registration procedure, grouped into Sales, Mortgages and Gifts. Every price, volume, median and yield on this site uses the Sales group only. Mortgage and gift registrations are never mixed into a price figure — their registered values follow different logic. They are published separately as registration activity (see Mortgages & gifts).

• One row, one property. A transaction here means one property within one registered sale procedure.
• Ready vs. off-plan follows the DLD’s registration type: Existing Properties (ready) vs. Off-Plan Properties. Off-plan values are contract prices registered before completion, often under payment plans — the two segments are therefore always separable in charts and filters.
• Property types — Land, Building, Unit (apartments) and Villa — are taken from the DLD’s own classification, unchanged.

Rent statistics are computed from officially registered Ejari rental contracts. The published population is deliberately narrower than the raw register —each exclusion below protects the medians from a known distortion:

• Apartments and villas only. Other residential types — labor camps above all, roughly a sixth of residential rows — rent at levels that would drag the market medians far from what a tenant or landlord would recognise. Commercial and industrial contracts are excluded entirely.
• Single-property contracts only. A contract can cover many properties at once (whole buildings, portfolios), and each of its rows repeats the contract-level total — a median computed over them would mix unit rents with building rents. Verified in the source: multi-property lines have a median registered amount more than twenty times the single-property median.
• New vs. renewal, always separated. New contracts price the current market; renewals are anchored to past contracts (and capped by the RERA rent calculator), so they lag by design. Headline rent figures use new contracts; renewals are shown alongside, labelled.
• Timing. Contracts are dated by their start date. Recent months are refreshed weekly from the DLD open-data gateway and replaced whole — the gateway publishes no contract identifiers, so months are re-pulled in full rather than merged row by row. History before March 2026 is frozen from the bulk archive.
• Rent per square foot. For apartments we also publish a rent per square foot: the median, across new single-property apartment contracts in the trailing twelve months, of each contract’s annual rent divided by its registered floor area. It is a per-contract ratio first, then a median, not the headline rent divided by a typical size. Apartments only — per-foot rates blend two different scales once villas are mixed in. Floor area is canonical in square metres and converted to square feet for display. Because smaller units rent at a higher rate per foot, the figure is reported by floor-area band as well as citywide and per area. A small share of contracts carry an implausible registered floor area (a few square metres, or thousands); these leave the rent amount itself valid, so they stay in the amount-based figures but are kept out of the per-foot ratio, which is published only within an absolute sanity band. The very largest size band (above 250 m²) is dominated by such mis-registrations and its per-foot figure is therefore suppressed.

Alongside sales, the DLD registers mortgages and gifts. We publish these as a measure of registration activity — counts, the registered mortgage amount and gift value — on a dedicated page, never folded into any sale price, volume, median or yield.

• A mortgage amount is a loan, not a price. The registered figure on a mortgage is the mortgage amount — across new mortgage registrations it runs at roughly seven-tenths of the comparable sale price and clusters on round figures, the signature of a negotiated loan rather than a valuation. We label it as a registered loan amount and never present it as a property price.
• Loan-to-value cannot be derived. No field links a mortgage to the sale of the same property, so a per-property loan-to-value ratio is not computable from this data and is never shown. The mortgage-to-sale figure we publish is a ratio of registration counts: new-mortgage registrations divided by sale registrations (a financing-activity proxy), not a share of buyers who used a mortgage. Modifications and transfers re-register existing loans, so they are excluded from the ratio's numerator.
• Mortgage types are separated. The headline counts new mortgage registrations; modifications, transfers and the development, portfolio and lease-finance procedures are shown as their own classes so a handful of very large corporate registrations never distort the typical figure. All amounts are medians with an inter-quartile range, never averages.
• Gifts are valuations, not sales. A gift is an ownership transfer; its registered figure is the valuation of the gifted property, so we report gift counts and that valuation, and never call it a price.
• Timing & history. Mortgage and gift rows are dated by registration, carry the same quality checks as sales (amount floor, plausible date) and the same recent-month under-count caveat, and are refreshed weekly from the DLD gateway on top of the historical archive.

Published statistics are computed only from rows that pass all of the checks below. Flagged rows remain in the database for audit but never enter a published number. As of June 2026 roughly 2% of source sales rows are excluded by these checks.

The headline statistic everywhere is the median: half of the registered sales were cheaper, half more expensive. Dubai’s market regularly produces record-setting penthouses and bulk deals; a median barely moves when they land, an average jumps. Averages are therefore never published.

• Spread is shown, not hidden. Where space allows, medians are accompanied by the interquartile range (25th–75th percentile) — the band that contains the middle half of all sales.
• Trailing windows are computed from raw rows. 12-month KPIs and similar stats are computed across all individual transactions in the window — never by combining monthly medians, because medians are not additive.
• Complete months only. Monthly charts exclude the month still being collected; a partial month would render a misleading dip at the end of every chart.
• Year-over-year compares the trailing 12-month window with the 12 months immediately before it, on the same population and rules.
• Price per sqft is a built-property metric. Per-square-foot medians, bands and trends cover built residential property — apartments and villas — only. Land trades on plot area (a single island spans tens of thousands of m², so its price per foot is a fraction of a built unit’s) and whole-building deals are wholesale, so both are excluded from psf; transaction counts still include every registered sale.

The price-momentum board on the market overview ranks areas by the year-on-year change in their median price per sqft — the trailing 12 months against the 12 months immediately before, on the same outlier-trimmed rules as every other median.

• Ready resale only. Off-plan launches register in waves at developer pricing, so an all-sales area median lurches when a new tower lands — a change in what sold, not in what a comparable home is worth. Momentum is therefore computed on ready (existing) apartment and villa sales only, the same restriction the yields methodology applies to its price side.
• Both windows must be liquid. A change between two medians is only meaningful when each side rests on real volume. An area is ranked only if it registered at least 50 ready sales in both the current and the prior 12-month window — a higher bar than the usual low-sample line, because a list of extremes attracts the noisiest areas.
• Anchored to the data, not the calendar. Both windows end at the latest registered-sale date in the dataset, so the comparison is never skewed by a still-incomplete current month.

A median of three sales is an anecdote, not a statistic. Every segment-level figure (area × month × property type and similar cuts) passes a sample-size guard before it is shown:

Suppressed segments are never backfilled with citywide values under a local label — a number labelled “Palm Jumeirah” is computed from Palm Jumeirah sales or not shown at all.

The price index shown on this site is the Dubai Land Department’s official residential price index — a quarterly series rebased by the DLD (base period around 2020 = 100). The index level and its quarter-over-quarter and year-over-year changes are republished exactly as the DLD computes them; we do not modify, extend or recalculate the series.

The index is methodologically independent of the medians computed on this site: the two can disagree in any given period, and seeing both is the point — the index is the DLD’s official view of price movement, the medians describe the raw registered transactions.

Yields combine the two registers: rents from Ejari contracts (the population defined in Rents), prices from registered sales. The formula:

gross yield (%) = median annual rent (new contracts, trailing 12M) ÷ median sale price (same segment, trailing 12M) × 100

• Same segment on both sides. Rent and price medians come from the same area × property type segment — apartment rents against apartment sale prices, villas against villas, each window anchored at its own dataset’s latest data. A yield is never computed from a local rent and a citywide price, or vice versa.
• Publication threshold. A yield is published only when the segment has at least 15 rent contracts and at least 15 sales in the window; below 30 of either it carries a lower-confidence badge.
• New contracts only. Renewal rents lag the market; the rent side uses newly signed contracts. Renewals are shown separately in the Rents section.
• Ready sales only. The price side uses completed (existing) properties. An off-plan unit cannot be rented, and off-plan contract prices reflect payment plans — mixing them in would inflate yields.
• Sanity band: 2–15%. An area-level yield assumes the rented and the sold stock are comparable. In a few mixed areas they are not — for example where the recent ready sales are hotel-apartment studios while the rental stock is family apartments — and the ratio lands far outside any realistic yield. Such values are suppressed as stock mismatch rather than published as the “best yield in Dubai”.
• Yield by size, not bedroom count. Where yields are broken down by unit size, the segment is the registered floor area (a band in m²), used as a durable proxy for bedroom count. The official rent feed that carries true bedroom labels is not available for the most recent contracts, whereas floor area is present on every record — so size bands keep the breakdown complete rather than letting it thin out over time. Size is an approximation of bedrooms, not an exact count.
• Gross means gross. Service charges, vacancy, agent fees and transaction costs are not deducted. Net returns are materially lower and depend on the specific property.

The stock pages count the existing inventory of registered units and buildings — the supply side — from the DLD units and buildings registers. This is distinct from the flow of sales and rents: a unit is counted because it exists, not because it traded.

• Bedroom mix is the register’s own room count. For residential apartments (the register’s Flat sub-type) we read the recorded room count — “Studio”, “1 B/R”, “2 B/R” … — and group four bedrooms and above together. This is a stock-side figure (how many one-bedroom apartments exist), not a count of deals. Staff and labour rooms and non-residential units (shops, offices, hotel rooms) carry no bedroom and are excluded from the mix while still counted in the unit total.
• Freehold share. The share of apartments flagged freehold in the register, citywide and per area. It describes the ownership-type composition of the existing stock, not the terms of any sale.
• Buildings. Each villa registers as one building; multi-unit towers additionally record the apartments, shops and offices they contain. Counts of villas, towers and their contents come straight from the buildings register.
• Aggregate-only. The 2.4 million units and 244 thousand buildings are summarised per area during processing; only the per-area totals are stored, never the individual records. The registers change slowly and are refreshed on demand rather than weekly.

• Currency. All values are in UAE dirhams (AED) as registered. No currency conversion is applied.
• Floor area. The source registers areas in square metres; that is the canonical unit in the database. Pages display square feet by default (the market convention in Dubai) with 1 m² = 10.7639 sqft, converted from canonical values through a single conversion path.
• Rounding. Displayed values are rounded for readability; underlying computations use unrounded values.

The canonical area registry is the DLD areas lookup — numeric identifiers with cadastral names such as “Al Barsha South Fourth”. The DLD’s live gateway, however, publishes marketing and community names for the same places (“Jumeirah Village Circle”). The two vocabularies are reconciled with an alias table derived from records that carry both forms, taking the majority mapping.

• Display names are the common ones. Areas are shown by their widely-used marketing/English name where it differs from the cadastral one (Al Khairan First is shown as “Dubai Creek Harbour”, Marsa Dubai as “Dubai Marina”). The official DLD cadastral name stays the record of truth and appears on each area page. One DLD area can span several named communities (Dubai Marina also covers JBR, Bluewaters and Dubai Harbour); since prices are registered at the area level, we label the flagship name and do not split it further.
• Unresolved names stay honest. Records whose area cannot be confidently resolved remain in citywide totals under an “unknown area” bucket — currently a fraction of a percent of all rows. They never appear under a wrong area, and area pages exclude them.
• Native codes are stored as-is. Transaction groups, procedures, property types and registration types keep the DLD’s own identifiers; no parallel classification is invented.
• Map boundaries. The choropleth uses official Dubai Municipality community boundaries (via Dubai Pulse, CC BY 4.0), joined to DLD areas by community number; a few areas without a published municipality boundary fall back to OpenStreetMap (ODbL). Boundaries are geographic context only — never a source for any metric — and some DLD sub-areas have no published polygon and stay unshaded.

Source data is © Dubai Land Department / Dubai Data Establishment, published as open data under the Creative Commons Attribution 4.0 licence. Attribution appears on every figure and in the site footer; the full dataset registry with links is on Sources.

• Values are processed. Numbers shown here are cleaned, deduplicated and aggregated derivatives of the source data. For the official record of any individual transaction, refer to the Dubai Land Department.
• Independent platform. This site is not affiliated with, or endorsed by, the Government of Dubai, the Dubai Land Department or any government entity.
• No advice. Nothing on this site is investment advice, a valuation or a recommendation to transact.

Knowing the limits of a dataset is part of using it honestly. The ones that matter most here:

• Registration lag. Transactions appear when they are registered, which can trail the underlying agreement by days or weeks. Recent totals firm up over subsequent weekly updates.
• Off-plan prices embed payment terms. Registered off-plan values reflect contracts with payment plans and are not directly comparable with ready resales — compare segments through the ready/off-plan split, not across it.
• Bulk registrations happen. Portfolio deals can register many properties at once. Medians limit their pull on headline numbers, and thin segments are protected by the low-sample guard.
• Missing attributes. Some records lack room counts or usable floor areas; segment cuts that need those fields exclude such rows, and price-per-area stats are computed only where area is present.
• No listings by design. Asking prices, advertised inventory and time-on-market are out of scope — they come from commercial portals, not official records.