How Fast Is Roulette? Measured Rounds, Stakes and Hourly Cost
An hour of roulette costs about €50 in expected losses on a single-zero wheel. That figure holds whether you play at a live-dealer table or against software but it gets there by two completely different routes, and the route matters more than the total.
Most writing about roulette speed is guesswork, because the honest answer requires knowing how often real people actually bet, not how fast a wheel can spin. One study measured it: 25.1 million roulette bets placed by customers of a single European operator, logged bet by bet. The numbers below are its measurements and our arithmetic on top of them.
What was measured
Michael Auer and Mark Griffiths published a player-tracking study in the Journal of Gambling Studies covering 43,731 players and 763,490 sessions at one major European operator serving Germany, Austria, the UK, Poland and Slovenia. Players were 82.6% male with a mean age of 43. Roulette appears in two separate rows of its game table — live dealer and software — which is what makes the comparison below possible at all.
| Measured | Live-dealer roulette | Software roulette |
|---|---|---|
| Bets observed | 19,054,053 | 6,058,289 |
| Mean interval between bets | 54.22 seconds | 20.27 seconds |
| Mean stake per bet | €27.83 | €10.95 |
| Standard deviation of stake | €184.11 | €46.88 |
| Mean session length | 36.12 minutes | 48.05 minutes |
One thing to be precise about, because it changes the interpretation: the interval is between a player’s own bets, not between spins of the wheel. A live table running a 25-second cycle still shows a 54-second player interval if the player sits out every other round, arrives mid-shoe, or takes a moment to decide. The measurement is of behaviour, not of equipment.
From seconds to rounds per hour
Divide an hour by the measured interval and the two products separate sharply.
| Per hour of play | Live dealer | Software |
|---|---|---|
| Rounds bet on | 66 | 178 |
| Total wagered | €1,848 | €1,945 |
| Expected loss, single zero (2.703%) | €50 | €53 |
| Expected loss, double zero (5.263%) | €97 | €102 |
The software player bets 2.7 times as often for 39% of the stake, and the two effects cancel almost exactly. Per hour, the products cost within a few euros of each other. Which is a tidy result, and also the least interesting thing on this page because people do not play for an hour. They play for a session.
The session is where they diverge
Software sessions in this data ran a third longer than live-dealer sessions: 48.05 minutes against 36.12. Combine that with the faster tempo and the gap opens up.
| Per average session | Live dealer | Software |
|---|---|---|
| Session length | 36.12 min | 48.05 min |
| Rounds bet on | about 40 | about 142 |
| Total wagered | €1,112 | €1,557 |
| Expected loss, single zero | €30 | €42 |
The product with the smaller stakes was the more expensive one. A €10.95 spin looks like a quarter of a €27.83 spin, and an average software session cost its player roughly 40% more in expected losses than an average live-dealer session.
This is the practical consequence of something covered on our house edge explainer: the edge attaches to each bet independently, so exposure is set by how many bets you place, not by how large they are or how long you sit there. Stake size scales what you win and lose. Round count scales how reliably the edge collects. Those are different levers, and the faster game pulls the second one without anyone noticing.
Why “mean stake” is not “typical stake”
Both stake figures above come with standard deviations far larger than the means themselves: €184.11 against a €27.83 mean for live dealer, €46.88 against €10.95 for software. For live roulette that is a spread 6.6 times the average.
A distribution that shaped is not a crowd of people betting roughly €28. It is a large number of modest bets alongside a small number of very large ones, and the mean sits well above what most bets actually were. So read the hourly figures as a description of the aggregate: this is what the average euro of turnover cost, across millions of bets. It is not a forecast for one person’s evening, and the person betting €5 a spin is not facing a €50 hour.
What the equipment advertises
Live-studio suppliers publish throughput figures for their fastest tables: one advertises 2,500 game rounds a day at 25 seconds a round, which works out to 144 rounds an hour if a player bet on every single one.
Measured live play came in at 66. The gap between 144 and 66 is the whole reason this page exists: a table’s cycle time tells you what is possible, and a player-tracking study tells you what happens. Roughly half the available rounds went unbet. Anyone quoting a supplier’s round count as a player’s exposure is overstating it by about a factor of two.
Four limits on these numbers
- The observation window was a lockdown. 27 November 2020 to 15 April 2021 covers a period when land-based venues were closed or heavily restricted across all five countries in the sample. Session lengths and tempo measured while the alternative was unavailable may not describe normal conditions. Of everything on this page, this is the caveat we would want tested first.
- One operator. The authors say so plainly: data from other operators could produce different results. A single operator’s product mix, table limits and customer base all shape these averages.
- The edge figures are theoretical, not the study’s. The study measured speed and stakes; it did not measure losses. Every euro figure on this page is the measured turnover multiplied by a theoretical house edge, which assumes the players were on single-zero or double-zero wheels as stated. The study does not report which wheels they played.
- Expected is not actual. Expected loss is the long-run average of a wide distribution. Across 25 million bets it is a reliable description of the aggregate. Across one 40-round session it is close to meaningless, and a majority of individual sessions will land nowhere near it in either direction.
Sources and method
Measured bet counts, intervals, stakes, standard deviations and session lengths: Michael Auer and Mark D. Griffiths, “The relationship between structural characteristics and gambling behaviour: An online gambling player tracking study”, Journal of Gambling Studies 39(1), 265–279 (2023), Table 1. Advertised studio throughput: live-supplier product documentation, retrieved 4 October 2026.
Rounds per hour, totals wagered and all expected-loss figures are our own arithmetic on the study’s reported values, not separate published statistics. House edge percentages of 2.703% and 5.263% are derived from the equal-pocket model and stated payouts. Stakes are in euros as the study reports them; the operator serves five countries, so some currency conversion is implied in the source and we have not adjusted it. Data as of 4 October 2026.