Slot RTP Explained: Theoretical Return, Observed Return and Sample Size
By Smiling Ghost · 4 October 2026 · Slot Mathematics
RTP (return to player) is the most quoted number in slot marketing and the most misread. People use the word for three different things: the return a game is designed to pay, the return a simulation measured, and the return that actually happened in someone’s session. This article separates the three and shows, with a worked example, how much sample size matters.
About the numbers. Everything below comes from a small illustrative model, “Game A”, that we built for this article. It is not one of our games. Exact values are calculated from its paytable; session results come from a simulation in Python/NumPy with a fixed random seed, so anyone can reproduce them.
1. Theoretical RTP: what the maths says
A slot’s theoretical RTP is the sum of every possible payout multiplied by its probability. For Game A:
| Outcome | Pays | Happens | Adds to RTP |
|---|---|---|---|
| No win | 0 | 69.97% of spins | 0 |
| Small win | 0.5x | 15% of spins | 7.5% |
| Win | 2x | 10% of spins | 20.0% |
| Good win | 5x | 4% of spins | 20.0% |
| Big win | 20x | 1% of spins | 20.0% |
| Top prize | 1,000x | 1 in 3,636 | 27.5% |
| Theoretical RTP | 95.00% | ||
Notice the last row. More than a quarter of the game’s return, 27.5 of the 95 percentage points, comes from a prize that lands about once in 3,636 spins. That single fact explains most of what follows.
2. Observed RTP: what actually happened
Observed RTP is total winnings divided by total bets over some number of spins. It wobbles around the theoretical value, and the size of the wobble depends on two things: how spread out the game’s results are, and how many spins you look at.
The spread is measured by the standard deviation of a single spin. For Game A it is 16.72 times the bet. The typical error of the observed RTP shrinks with the square root of the number of spins:
| Spins observed | Typical error (1 standard deviation) |
|---|---|
| 100 | ±167 percentage points |
| 1,000 | ±52.9 points |
| 10,000 | ±16.7 points |
| 100,000 | ±5.3 points |
| 1,000,000 | ±1.67 points |
| 100,000,000 | ±0.17 points |
To pin Game A’s RTP down to within half a percentage point, you need about 11 million spins. Anything smaller is mostly noise.
3. Sessions: why the average is not the typical result
We simulated 10,000 sessions of each length:
| Session length | Average return | Median return | Sessions that ended ahead | Middle 90% of sessions |
|---|---|---|---|---|
| 200 spins | 94.5% | 67.0% | 8.2% | 44% to 543% |
| 1,000 spins | 94.9% | 70.5% | 24.0% | 57% to 177% |
| 10,000 spins | 94.8% | 94.4% | 32.2% | 69% to 125% |
The average stays close to 95% at every length, as it should. The median, the result of a typical session, is far lower in short sessions: two thirds of the stake back after 200 spins. The reason is the top prize. In 200 spins there is a 94.6% chance that the 1,000x prize never lands, so most short sessions run on the remaining 67.5% of the return. A few lucky sessions that do hit it pull the average up.
4. What this means in practice
- RTP describes the game, not a session. A 95% RTP does not mean you get 95 back from 100. It means that over a very large number of rounds the game pays back about 95% of what is wagered.
- Short-term results say almost nothing about RTP. A few thousand spins of data can sit tens of points away from the true value without anything being wrong.
- Look at where the RTP comes from. Two games with the same RTP can feel completely different, depending on how much of the return sits in rare prizes. We cover this in Slot Volatility vs Hit Frequency.
How we handle RTP in our own games
Our games run on Stake Engine. There the set of possible outcomes of each game mode is generated in advance and stored with weights, so the RTP of the game is calculated exactly from that stored table, not estimated. We publish it in the game rules for every mode. More on our process on the Game Math page.
