Why a deck needs shuffling at all
Card probabilities assume a shuffled deck: any order of the 52 cards is as likely as any other. A brand-new deck, or one sorted after a game, is far from that. Shuffling is the process meant to remove the order, and the sensible question is how many shuffles it takes before the leftover order no longer matters for practical purposes.
There are 52 factorial possible orderings, about 8.07 x 10^67. A truly uniform shuffle would make each of them equally likely. No physical shuffle reaches that exactly. What can be studied is how close a method gets, and how that closeness improves with repetition.
What a single riffle does
A riffle shuffle splits the deck into two packets of roughly equal size and interleaves them, so cards drop alternately from each packet in an irregular pattern. Importantly, the cards within each packet keep their relative order. After one riffle, the original deck can be recovered by picking out two interleaved increasing sequences.
This limits the result. For a deck of n cards, a single riffle can produce only 2^n minus n distinct orderings. For 52 cards that is a minuscule share of the 52! possibilities. One shuffle leaves strong traces of the starting order, such as long stretches of cards that were adjacent before and are still in the same relative order.
The mathematical model
To study the question, mathematicians needed a precise description of a riffle. The standard one is the Gilbert-Shannon-Reeds model. It cuts the deck at a random point chosen from a binomial distribution (so cuts near the middle are most likely), then drops cards from the two packets one at a time, with the chance of taking from a packet proportional to how many cards it has left. It is a model, not a measurement of any person’s hands, but it captures the main features of a casual riffle and is mathematically tractable.
Within this model, you can ask how far the deck’s arrangement is from uniform after k shuffles. A standard way to measure the gap is total variation distance: roughly, the largest difference between the model’s probability of an event and the probability under a perfectly uniform deck. A value near 1 means the deck is easy to distinguish from random, and a value near 0 means it is hard.
The result about seven
In a well-known 1992 paper, Dave Bayer and Persi Diaconis analysed riffle shuffles for a 52-card deck. Their main finding is often summarised as “seven shuffles are enough.” The distance stays close to its maximum for the first few shuffles, then drops quickly around the fifth to the seventh, and after that it keeps shrinking, roughly halving with each extra shuffle. After about seven, the deck is generally considered close enough to random for ordinary card games.
The word “close” matters, and the claim is easy to overstate. Seven shuffles do not make the deck exactly uniform. The result depends on the measure used, and a different measure of closeness could give a somewhat different figure. It also assumes the Gilbert-Shannon-Reeds model, so it applies to a particular idealised style of shuffling. A person who riffles unevenly, or who cuts clumsily, may need more.
The researchers also noted that the answer depends on deck size: the number of shuffles needed grows slowly with the number of cards. Doubling the deck does not double the shuffles; the growth is roughly logarithmic in the number of cards, which is part of why seven is enough for 52 while a much larger deck needs only a few more.
Other ways to shuffle
Other methods behave differently:
- Overhand shuffles move small clumps from one hand to the other. They mix far more slowly than riffles, and a very large number of them is generally needed.
- Cutting the deck by itself only rotates it; the cyclic order of the cards is unchanged.
- Perfect faro shuffles, where the deck is split exactly in half and interleaved exactly, are not random at all. Eight perfect out-shuffles in a row bring a 52-card deck back to its original order, which is a nice demonstration of how a deterministic process can look like mixing without being random.
- Machine shuffling and shuffling several decks change the picture again, and the research above does not directly tell you how a given machine performs.
- Software shuffles replace hands with pseudo-random generators, so the quality of the result depends on the algorithm and the generator used rather than on a count of shuffles.
Why it matters
The calculation behind every card probability, from five-card poker hands to bridge deals, assumes a uniform random order. Imperfect shuffling can in principle leave patterns, and the research shows that a few careless shuffles do leave real structure behind. In casual play, with several riffles and some cutting, the practical effect on a card game is usually negligible, but the “does not matter” claim rests on mixing enough.
It is worth being careful in the other direction too. A mathematically imperfect shuffle does not give an ordinary player a way to predict cards, and it does not create a dependable advantage in a game where the structure of the game takes a margin. Games with a house edge lose money for players over time, and the odds of a particular hand do not change because of anything that happened in the previous shuffle.
Understanding shuffling is a way to see why probability models treat a deck as random and how reasonable that assumption is. It also illustrates a general point about randomness: a small number of well-understood steps can bring a system near uniform, while a large number of poorly mixing steps may not.