What dice control claims, in plain terms
The theory is mechanical. If you start two dice in exactly the same position and throw them in exactly the same way, they should tumble the same way and land the same way. Real throws are never identical, so proponents claim less: that a practiced shooter can keep the dice rotating on one axis often enough to shift the odds.
Two separate ideas hide under the name. Dice setting is arranging the faces before the throw, which anyone can do. A controlled throw is the harder claim: that you can make the dice fly together, spin backward on the same axis and die at the back wall without scrambling. Setting without that throw changes nothing at all.
The idea is old. John Scarne described controlled rolls in Scarne on Dice in 1974, along with ways to detect them. Jerry Patterson and Chris Pawlicki launched a course on setting and controlling dice in 1997, and Stanford Wong wrote Wong on Dice in 2005. Casinos have known about the claims for decades.
Dice setting craps players use, and why the axis matters
A die has three axes, each running through a pair of opposite faces. If a die spins cleanly on one axis, only the four faces around it can come up. With both dice on axis, 36 possible combinations shrink to 16, which is the whole premise. The Wizard of Odds counts 84 distinct ways to set a pair before throwing.
Do the arithmetic yourself. Opposite faces always add to 7. If both dice spin on the same axis, 4 of the 16 on-axis combinations are sevens, which is 25%. Put the two dice on different axes and only 2 of 16 are sevens, 12.5%. That difference is what every craps dice setting chart is trying to exploit.
| Throw | Possible combinations | Sevens among them | Chance of a 7 |
|---|---|---|---|
| Fully random | 36 | 6 | 16.67% |
| Both dice on the same axis | 16 | 4 | 25.00% |
| Dice on different axes | 16 | 2 | 12.50% |
Hence the set names. A sevens set, such as the one shooters call the 3-V, is for the come-out. A hardways set, which the Wizard of Odds analysis ranks as the best for making a point before a 7, is meant for the point stage. It works only if the axis survives the wall.
What the math says about the best dice set for craps
Random throwing keeps both dice on axis about 44% of the time, by the Wizard of Odds’ reckoning. Even then the two dice rarely rotate in step. Shackleford’s model counts a throw where one die turns an extra half-turn against the other, and for a random shooter that and a perfectly matched landing are equally likely, at 11.11% each.
His hypothetical tables show the temptation. If a shooter converted just 1% of those mismatched throws into matched ones, the pass line with 3-4-5x odds would move from a 0.374% house edge to a slight player advantage. The tables assume control works. Shackleford writes that he remains skeptical, and that he would allow setting if he ran a casino.
The best dice set for craps, if control exists, is therefore the Hard Ways Set #1 during the point and the Sevens Set #1 on the come-out, by his analysis. If control does not exist, every set is equal, the sevens-to-rolls ratio is 6, and the craps odds you already know apply to every throw you make.
Is dice control real? The tests so far
In July 2004 the Wizard of Odds reported a small shooting test with Beau Parker and others at the Bellagio and the Westin. Come-out winners came in at 11 against an expected 8.22, but the rest of the results were close to random, and the author called the sample too small for any firm conclusion.
The August 2004 test is the famous one. Wong and a partner threw 500 rolls with a line of 79.5 sevens, where 83.33 would be expected. They threw 74. That sounds convincing, but a purely random shooter throws 74 or fewer sevens in 500 rolls about 14% of the time, by our own calculation, roughly one test in seven.
The most careful work is academic. Donald Smith and Robert Scott, backed by a Monmouth University grant, first calculated the control needed to break even, publishing in the UNLV Gaming Research and Review Journal in 2018. In 2020 they built a throwing machine, which they named Lucky Lil, to mimic the backspin and on-axis flight of expert shooters.
They filmed its throws on a six-foot craps table with a high-speed 4K camera, calibrated it, and recorded 7,557 throws, testing the results with chi-squared statistics. The paper is titled Pair-a-Dice Lost. As it has been summarised, the machine failed to achieve a significant advantage in any of the setups tested, a poor sign for human hands.
How casinos stop controlled throws
Casino tables are built against the theory. The inside of the far wall is lined with rubber pyramids, and dealers insist that the dice hit it, which is exactly the collision that scrambles an axis. The shooter must use one hand. A throw that stops short of the middle of the table can be called no roll and thrown again.
That is why the practical advice on how to throw dice in craps is so plain. Hold them loosely, keep them low, and send them firmly enough to hit the back wall without leaving the table. Our craps etiquette guide covers the table manners around the throw, from keeping your hands clear to when you may set.
Test your own throw with real numbers
The only honest way to judge your own dice control is to count. Record every roll, thousands of them, and compare your sevens with the one-in-six you would get at random. A ratio of 6.3 rolls per seven needs about 9,000 rolls to stand clear of variance at two standard deviations; a ratio of 7 needs about 1,000.
Our roll log is built for that. Enter your totals as you throw, and it tracks your sevens-to-rolls ratio and runs a fairness test against the expected distribution. Log a few hundred rolls at home first. If your 7 count still sits near one in six after 1,000, you have your answer before a casino gets it.
Until the numbers say otherwise, bet as if every throw is random, because on the evidence it is. That means the pass line with odds, placing the 6 and 8, and staying away from the hardways and one-roll bets that dice setters love. Set the dice if you enjoy it; the back wall decides the rest.