How Slot Machines Work: RNG, Reels and Payout Odds
Most slot machines in a U.S. casino do not spin reels at all. They run a random number generator that picks an outcome from millions of possible combinations, then animate a reel display to match it, and the payout odds are set in software before the first bet is ever placed. The reels are theater; the math is the machine.
That distinction matters more than it sounds. Players who understand it stop looking for patterns in the last ten spins, stop believing a machine is "due," and start asking the questions that actually change their outcome: what is the RTP, how volatile is the game, and how many spins does it take for the advertised percentage to mean anything.
What the RNG actually does
Every modern slot runs a pseudorandom number generator, a deterministic algorithm seeded with some starting value. On each spin the machine draws a number, maps it to a position on each virtual reel, and resolves the outcome in milliseconds. The animation that follows is a replay of a result that already exists.
This is where most player intuition breaks down. Nothing about the previous spin constrains the next one. A machine that has paid nothing for 400 spins is not warmer than it was at spin one. The only thing that changes over time is your bankroll.
The engineering underneath is less glamorous than the marketing. RNG seeds in mobile slot apps have a documented tendency to repeat under heavy spin volume, a problem engineers have traced to seeding and cycle length in constrained environments — the same class of issue covered in why slot RNG seeds reuse after 40,000 mobile spins. On regulated casino floors, the RNG is a sealed, certified component, and repeat cycles are one of the things labs test for.
The 1989 rule that made this auditable
Nevada got there early. In July 1989 the Nevada Gaming Commission approved amendments to Regulation 14 setting standards for how game outcomes must be selected — statistical testing at a 95% confidence level, requirements around the independence of outcomes, and a rule that the display must accurately represent the result. That last clause is the legal reason a machine cannot show you three cherries and pay you for two.
Regulators elsewhere followed with their own frameworks, but the core principle is now standard: the outcome is determined by the RNG, the display must reflect it, and the whole thing has to survive independent testing before it reaches a floor.
Why the reels lie about your odds
A three-reel machine with 20 symbols per reel looks like it has 20 × 20 × 20, or 8,000 possible outcomes. It usually doesn't. According to the Missouri Gaming Commission, a game program may use 256 virtual stops per reel even though only 20 symbols are visible — which gives 256 × 256 × 256, or 16,777,216 combinations. The chance of hitting the jackpot on such a machine is 1 in 16,777,216, not 1 in 8,000.
The visible symbols are a sample of the virtual reel. Blank stops sit between the symbols you can see, and the weighting of those stops is what sets the true probability of each symbol landing on the payline. This is why two machines with identical artwork can have wildly different payout profiles.
On video slots the same logic applies with more moving parts. A five-reel game with 30 to 50 stops per reel can have hundreds of millions of combinations, and the paytable is calibrated against that full space. The screen shows you five symbols; the math sees the whole set.
| What players see | What the machine uses |
|---|---|
| 20 symbols per reel | 256 virtual stops per reel |
| 8,000 visible combinations | 16,777,216 actual combinations |
| "It's been cold all night" | Independent draws, no memory |
RTP, hit frequency and the number that actually matters
Return to player is the long-run percentage of all wagers a game returns to players. A 96% RTP game keeps 4 cents of every dollar wagered, on average, over a very large number of spins. That is the house edge, and it is not a promise about your session.
The measurement window is the part most players miss. In Great Britain there is no minimum RTP requirement for gaming machines at all, and the Gambling Commission notes that average RTP is generally measured over 10,000 or 100,000 games for compensated machines, and over more games than that for random machines, depending on category. Read that again: the percentage on the glass is an average across a sample that can run into the hundreds of thousands of spins. Your 200 spins are a rounding error inside it.
Hit frequency is a separate number and often more useful. A game with 96% RTP and a 20% hit frequency feels very different from one with 96% RTP and a 45% hit frequency, even though the long-run return is identical. The first pays less often but larger; the second pays small amounts constantly. Neither is better — they suit different bankrolls and different tolerances for sitting through dry spells.
Compensated versus random machines
The distinction the UK regulator draws is worth knowing. Compensated machines adjust outcomes based on prior results within a cycle, which is why their RTP can be measured over a defined 10,000- or 100,000-game window. Random machines, the standard in U.S. casinos and most regulated markets, draw each outcome independently. Their RTP converges only over a much longer horizon.
If you are playing a random machine, the concept of a "cycle" is meaningless. There is no point at which the machine owes you a payout.
What the floor data says about where the money goes
Slots are not a side business for casinos. They are the business.
Nationwide, slot machines produced $91 billion in revenue in 2025 — half of the $182 billion reported by the U.S. casino industry, according to figures from the Gaming Control Board and a Southern California gaming research firm. In Nevada specifically, slots accounted for $10.7 billion, nearly 68% of the state's $15.8 billion in total gaming revenue for 2025.
Those are gross gaming revenue figures, meaning player losses after payouts. They are also the clearest argument for why the RTP math is not abstract. A machine that returns 90% versus one that returns 96% is a six-point swing on every dollar cycled through it, and at national volume that gap is measured in billions.
The per-machine economics are equally blunt. An industry estimate puts slot machine profit at around $100,000 per machine per year in player losses. A player who spends three hours one evening a week at $1 per spin would lose roughly $1,000 a month, or about $12,000 a year, on average — a figure reported by Kevin Harrigan in The Conversation in 2014. At $1 spins that is a lot of lever pulls, but it is not an unusual recreational pattern.
None of this requires the machine to be rigged. It requires only that the house edge compound across volume, which it does.
Why session RTP drifts from advertised RTP
Advertised RTP assumes a clean, uninterrupted sample. Real sessions are messier. Players speed up when they are losing, slow down when they are winning, switch machines after a bad run, and cash out at arbitrary points. Each of those decisions reshapes the sample you personally experience.
There is a body of work on this blog about how environmental factors skew measured payouts — cashier shift changes trimming slot payouts by 8% is one example of how operational timing shows up in the data. Whether or not you buy the specific mechanism, the underlying point holds: the RTP you experience in a two-hour session is a small, noisy sample, and small samples wander.
How to read a slot machine before you play
The regulatory disclosures are usually there, just not where the marketing wants you to look.
Find the RTP. In many jurisdictions it must be published — on the game's information screen, in the paytable, or on the manufacturer's site. If a casino will not tell you the RTP of a specific game, that is information in itself. A 92% game and a 97% game can sit side by side on the same floor with identical themes.
Check the volatility. Low-volatility games pay smaller amounts more often; high-volatility games pay rarely but potentially large. High volatility with a small bankroll is how sessions end in twenty minutes.
Ignore the near-miss. A jackpot symbol one stop above the payline is not a signal. On a virtual reel with 256 stops, the near-miss is a designed visual event, and regulators in most mature markets require that it not misrepresent the actual probability.
Set the loss limit before you sit down. The $12,000-a-year figure above assumes discipline about stake size, and most players do not have it once a machine is paying. Tools like a daily loss cap set at $200, which blocks 1 in 3 weekend deposits, exist precisely because willpower is unreliable at the machine.
If you play online rather than on a floor, the same questions apply, and the data trail is longer. Software-side issues that never surface in a land-based cabinet — stream timing, frame sync, backpressure — can distort what a mobile player sees, as in this analysis of Node.js stream backpressure corrupting slot reel sync at 60fps. A display that stutters is not changing the RTP, but it is changing what you think you saw.
The question the industry has not answered
The mechanics are settled. RNG picks the outcome, virtual reels set the probability, RTP sets the long-run return, and the display is a rendering of a result that already happened. Every regulated machine in the U.S. and most major markets runs on that architecture, tested to a 95% confidence standard and audited against the display requirement Nevada wrote into Regulation 14 in 1989.
What is not settled is disclosure. RTP is published in some markets and buried in others. Volatility ratings are inconsistently labeled. Hit frequency is rarely shown at all. A player can walk onto a floor with two machines that look identical, play the same game at the same stake, and face house edges that differ by five points — with no obligation in many jurisdictions to make that visible.
The UK regulator's own guidance notes there is no minimum RTP for gaming machines there. If a jurisdiction will not set a floor, the least it could do is require the number to be legible. Whether that happens — and whether players would change their behavior if it did — is the open question the next round of gambling regulation will have to answer.