What casinos actually keep
Single-zero roulette has a theoretical house edge of 2.703%. That number comes from the geometry of the wheel, not from any casino’s accounts. The obvious question is whether it holds up when real people bet real money — and for most of gambling’s history, nobody outside the operators could check.
Spain’s gambling regulator publishes both halves of the equation. Between 2021 and 2025, licensed operators there took €42.2 billion in roulette wagers and kept €1.02 billion of it. That works out to a realised margin of 2.42% across five years — close to the theoretical 2.703%, and consistently a little below it.
This page sets the theory next to the measurement, explains the four different percentages that get confused with each other, and is honest about the one thing the data does not explain.
The theory: three wheels, three edges
The house edge in roulette comes from a single design decision: the wheel has more pockets than the payouts account for. A single number pays 35 to 1, but on a European wheel there are 37 pockets. That gap is the entire business model.
| Wheel | Pockets | House edge | Return to player (RTP) |
|---|---|---|---|
| American, double zero | 38 | 5.263% | 94.737% |
| European, single zero | 37 | 2.703% | 97.297% |
| French, la partage on qualifying even-money bets | 37 | 1.351% | 98.649% |
Two cautions before these numbers travel any further. First, the la partage advantage is narrow: it halves the loss on even-money bets when zero lands, and does nothing at all for a straight-up number. A French table is not a 1.351% game; it is a 2.703% game with one rule that improves one family of bets.
Second, “American roulette” is not a reliable label for a double-zero wheel. The UK Gambling Commission uses the term as a reporting category for physical roulette tables regardless of their actual configuration. If you want to know the edge you are facing, count the pockets. Do not read the name of the game.
The measurement: €42.2 billion in Spanish wagers
Spain’s Dirección General de Ordenación del Juego is unusual in reporting both the amount wagered and the amount the operator kept, split between live-dealer and software roulette. Divide one by the other and you get the margin the house actually realised.
| Year | Wagered | Operator kept (GGR) | Realised margin |
|---|---|---|---|
| 2021 | €5.97bn | €142.6m | 2.39% |
| 2022 | €7.15bn | €176.5m | 2.47% |
| 2023 | €8.88bn | €216.1m | 2.43% |
| 2024 | €9.13bn | €234.5m | 2.57% |
| 2025 | €11.08bn | €251.4m | 2.27% |
| Five-year total | €42.22bn | €1,021m | 2.42% |
The headline figure is the stability. Five years, five different volumes, and the margin never leaves a band between 2.27% and 2.57%. Nothing about online roulette turns out to be mysterious: the wheel’s arithmetic shows up in a national regulator’s accounts, at scale, year after year.
Live-dealer roulette and software roulette landed in the same place, which is worth noting because they are very different products. Live tables carried €37.0 billion of that €42.2 billion total and grew their revenue by 18.3% a year compound; software roulette stayed flat. The margins, though, track each other closely — between 2.24% and 2.58% on both sides across all five years.
Roulette next to blackjack and slots
The same regulator reports wagers and revenue for other games, which makes a rare apples-to-apples comparison possible. Same market, same year, same definitions.
| Game, Spain 2025 | Wagered | Operator kept | Realised margin |
|---|---|---|---|
| Roulette | €11.08bn | €251.4m | 2.27% |
| Blackjack | €0.96bn | €24.4m | 2.54% |
| Slots | €14.19bn | €617.1m | 4.35% |
Roulette was the cheapest of the three per euro staked in 2025, and it was not close on slots. That is a statement about the mathematics of each game, not advice about which to play — a lower margin applied to more rounds can cost you more than a higher margin applied to fewer, which is the subject of its own page.
Why is the realised margin below the theoretical edge?
Every year in the table comes in under 2.703%. We do not know why, and we would rather say so than invent a reason. What we can do is set out the candidate explanations, because one of them matters a great deal and the rest are housekeeping.
- Bet mix. French-rule tables are widely available in the Spanish market, and la partage drops the edge on even-money bets to 1.351%. A player base that favours red/black and odd/even would pull the blended margin below the single-zero figure on its own.
- Promotional credit. If bonus funds are counted in the wagered total but the resulting payouts are not netted the same way, the denominator grows faster than the numerator.
- What counts as a wager. Regulators differ on how recycled stakes, cancelled rounds and side bets are recorded. A definitional difference of a few percent in the denominator moves the margin more than any player behaviour does.
- Variance. Over €42 billion of turnover this is the least likely explanation. At that volume the law of large numbers is not a theory, it is a floor.
What this is not is evidence of an exploitable gap. A sector-wide average margin sitting below the single-zero edge tells you something about how bets are distributed and recorded across millions of players. It tells you nothing you can act on at a table, and anyone who presents it as a strategy is misreading it.
Four percentages that are not the same thing
Most of the confusion around roulette numbers comes from four figures that all look like “the house’s cut” and have four different denominators. Once you can see the denominators, the apparent contradictions disappear.
| Figure | What it divides | By what | Example |
|---|---|---|---|
| House edge | Expected loss per bet | Amount of that bet | 2.703% on a single-zero wheel |
| Return to player (RTP) | Expected return per bet | Amount of that bet | 97.297% — the same fact, stated from the other side |
| Realised margin | Revenue the operator kept | Everything wagered, including re-wagered winnings | 2.27% in Spain, 2025 |
| Hold | Revenue the casino kept | Cash and credit exchanged for chips at the table | 19% in Nevada, 2025 |
Hold is the one that trips everybody up
Nevada’s roulette tables reported a hold of 19% of drop in 2025, according to the Nevada Gaming Control Board. Seven times the theoretical edge. Nothing is wrong with either number.
Drop is what a player hands over at the table to buy chips. Those chips then get bet, won back, and bet again — often many times over before the player leaves. The edge applies to each of those bets; hold applies once, to the money that came in. A player who buys €100 in chips and places forty €20 bets has wagered €800, lost roughly €21.60 to the edge, and contributed to a hold figure calculated on the original €100.
Nevada’s own figures show the regional spread this produces: 18.64% on the Las Vegas Strip, 21.49% in Reno, 25.03% in Elko County. The wheels are identical. What differs is how long players stay and how often they recycle their chips.
For scale, Nevada’s 424 reported roulette tables across 96 locations won $407.7 million in 2025 — 10.33% of the state’s table-game win, but only 2.58% of its total gaming win. Which share you quote depends entirely on which denominator you pick, and both are correct.
The outlier worth knowing about
Not every machine with a wheel on it returns 97%. A study of Norway’s Multix terminals recorded an observed return to player of 84.88% on the roulette game, from logged play in January 2010.
That is a bespoke terminal product from a state monopoly operator, not a standard European wheel, and the figure is a measured outcome from one month rather than a published RTP. It belongs here as a boundary marker: the 2.703% edge is a property of a specific wheel with specific payouts. A game that calls itself roulette is not required to share them, and the only reliable way to know is to read its rules.
What this means at the table
Three things follow from the numbers above, and none of them is a system.
- Pocket count is the only choice that changes your odds. Moving from a double-zero wheel to a single-zero wheel halves the edge, from 5.263% to 2.703%. No betting pattern available to you will do anything comparable, because no betting pattern changes what the wheel pays.
- The edge compounds with rounds, not with time or with stake size. It applies to every bet independently. Two hundred small bets expose you to more expected loss than twenty large ones of the same total, which is why speed of play matters more than most players assume.
- Spain’s five years are the long run, and you are not playing it. A regulator averaging €8 billion a year lands within a quarter-point of the theoretical edge. One evening at a table is a sample of a few dozen spins, where the range of outcomes is enormous and the average is almost meaningless. The margin is what the operator can count on. It is not what will happen to you tonight.
Sources and method
Spanish wagering and revenue figures: DGOJ annual interactive data. Nevada hold, win, table counts and regional breakdown: Nevada Gaming Control Board, December 2025 monthly revenue report. Share of total and table-game win: UNLV Center for Gaming Research, Nevada game mix. Multix return to player: Leino et al. Reporting-category terminology: UK Gambling Commission definitions.
All realised margins, five-year totals and share figures on this page are our own arithmetic on the reported inputs, not separate published statistics. House edge and RTP figures are derived from the equal-pocket model and stated payouts. Monetary figures are in the currency the source reports; nothing on this page converts between currencies or sums across markets, because the underlying definitions differ. Data as of 4 October 2026.