Pick your denominations, from the penny game up to the hundred dollar machine. Same bankroll, same bet, same number of spins. The only thing that changes between them is the payback percentage. Run it and watch how little that number decides about the next forty minutes, and how completely it decides everything after that.
Every number on this page comes from the generator picked here. Real machines run a PRNG on purpose, because a regulator has to be able to seize it, re-seed it and replay it. You cannot audit atmospheric noise.
The way you will actually experience a slot machine: once, for about as long as a lunch break.
The 96% figure people quote for Las Vegas is real, but it is not a floor average anywhere. It is what nickel machines return on the Boulder Strip and what quarters return in North Las Vegas, both locals markets. The Strip floor averages 92.10% and a compacted tribal floor in Connecticut averages 91.85%. Those are close to the same number, which means the line that actually divides the country is tourists against regulars rather than Nevada against everywhere else.
One caveat travels with every figure on this page. Regulators report reel slots, video poker and video keno together inside each denomination, and video poker returns far more than a reel game. It is concentrated in nickels and quarters, which is why some of the ladders below bend rather than climb: on the Strip, quarters report 89.06% while nickels report 91.62%. A nickel reel slot is not paying 96% anywhere. A bank of nickel video poker sitting in the same denomination bucket is what lifts the average.
One session proves nothing either way. So run a few thousand and count who finishes on top.
If the machines were identical, pure luck would put each one in first place 12.5% of the time. Everything above that line is what the extra payback percentage actually buys you inside one session.
Here the payback percentage shows up precisely as advertised, in the average. Nobody gets to play the average. You play one session, once, and land somewhere in the spread.
The advertised number is a property of the machine's whole life, not a forecast for yours.
Payback percentage is a rate. What it's worth in dollars is decided by the size of the bet it's applied to.
Four sources, tested side by side on two separate questions. Does the output look random, and can you guess what comes next. Those are not the same question, and only one source here answers yes to the first and no to the second.
This is the part that separates statistical quality from unpredictability. Below, one Mersenne Twister runs as the machine. A second one watches 624 of its outputs, reverses the tempering step on each, loads the result as its own state, and then calls every subsequent number before the first one produces it.
This does not transfer to a cabinet on a casino floor. You would need 624 consecutive raw 32-bit values, and a machine gives you reel positions instead, from a generator that keeps cycling whether or not anyone is playing. It has still happened: a group working from reverse-engineered firmware on older Aristocrat machines timed spins off a predicted stream for several years before arrests in 2014. The countermeasure is not a better paytable, it's a generator whose next output is not a function of the ones you have already seen.
The chi-square column is the test most people assume is the whole job: does every value show up about as often as every other. The two good sources land somewhere around 63, which is what sixty-four bins of real noise looks like. The broken one comes back near zero, and that is worse than a high number, not better. A full-period generator visits every one of its states exactly once, so its histogram comes out flatter than chance could ever manage. Bins that even are a fingerprint, not a clean bill of health, which is why the check here fails in both directions.
The scatter plots are the other half. Each dot is one draw against the draw that followed it. Real noise fills the square. The broken generator collapses onto a handful of straight lines, because 65,536 states is small enough to see and each state determines the next exactly. Once you know two spins you know every spin after them. That is what certification standards are actually looking for when they test serial correlation and lattice structure instead of just counting bins.
MT19937 is the one worth sitting with. It passes both columns cleanly. Its period is 2^19937, which is a number with six thousand digits, and it was good enough to become the default generator in Python, Ruby, PHP, R, MATLAB and the C++ standard library. Nothing in any statistical test above catches it. It is still fully predictable, and the section below does it live.
In tribal casinos without a Class III compact, the cabinet that looks like a slot machine is legally a bingo game. Everything above this section stops applying.
The Indian Gaming Regulatory Act sorts games into three classes. Class III covers house-banked games, real slot machines included, and a tribe needs a negotiated compact with the state to offer them. Class II covers bingo and games similar to bingo, and needs no compact. So where a compact is absent or narrow, the floor fills with machines that satisfy the legal definition of bingo while looking and sounding exactly like the machines in section 01.
Underneath, a central server deals every player at the bank a bingo card and draws balls. Your card is checked for patterns, your prize is decided, and only then does the cabinet pick reel symbols that add up to the amount you already won. The reels are a readout, not a mechanism. There is no paytable behind them in the sense used everywhere else on this page.
| Class III slot | Class II bingo machine | |
|---|---|---|
| What it legally is | A slot machine | A bingo game with an electronic aid |
| Needs a state compact | Yes | No |
| What decides your outcome | Your spin, independently | A ball draw shared with the bank |
| Who you are playing against | The house | The other players, plus the rake |
| What the reels do | Determine the result | Narrate a result already decided |
| Primary regulator | State gaming board, per compact | National Indian Gaming Commission |
| Minimum payback | Set by compact or state law | No federal minimum |
| Payback published | Often, by denomination | Generally not |
Worth being clear about what this does and does not say. Plenty of tribal casinos run genuine Class III floors under compact, and those machines are mathematically identical to the ones in Nevada, so sections 01 through 04 apply to them unchanged. Class II is what fills the gap where a compact is absent or narrow. The tell is on the screen: a Class II cabinet has to show you the bingo card it is playing, usually shrunk into a corner, and often a daub button that does nothing but let you press something.