Is card counting illegal? No. Here is what happens instead.

In the United States, counting cards is not a crime in any state. You are watching cards that were dealt in front of you and doing arithmetic about the ones that are left. There is no statute against it because there is nothing to prohibit: no device, no marked cards, no altered outcome, no confederate. It is thinking, and thinking about a game you are permitted to play is not a criminal act. This is general information about how the law has worked, not legal advice about your situation.

The confusion comes from somewhere real. Casinos have spent decades letting the word "cheating" sit next to the word "counting" in public. Being unwelcome and being unlawful are different things, and the gap between them is where everything interesting lives.

Why it isn't cheating, in the language of the statute

Nevada defines cheating (NRS 465.015) as altering the elements of chance, the method of selection, or the criteria that determine the result of a game. Many states use a similar shape, though the wording is not uniform and you should read your own state's rather than Nevada's. Read it against what a counter does. The house shuffles, a player cuts, the dealer deals by written procedure. The counter changes their own bet size and, sometimes, their own hit-or-stand decision, both of which the house offers every player at the table. Nothing about the cards, the order, or the payout is touched.

That is the entire legal argument, and it is why prosecutors do not bring these cases: there is no element to prove.

What the casino may do instead, and it is plenty

Legal does not mean protected. The right to exclude is state law and it varies. In most jurisdictions a casino floor is private property, and a licensee may refuse service for nearly any reason that is not a protected-class one under federal or state public-accommodation law. "You are too good at this" qualifies. New Jersey is the exception, below; tribal rooms and non-US regulators run on other rules again.

The escalation usually runs:

• Shuffle-up. The dealer reshuffles early, whenever you raise. It removes the deep-shoe hands the count exists to find.
• No mid-shoe entry. You may play, but only from the top of a shoe.
• Flat betting. You may keep playing blackjack at exactly one bet size. Cheerful, and fatal: one bet size removes the mechanism counting works through.
• The back-off. A polite request to play anything else.
• The trespass notice. Read aloud, sometimes in writing, sometimes with a photograph taken. After that, returning is criminal trespass in most states — a real crime, unlike the counting that provoked it — though what makes a notice effective, and how long it stands, is state law and differs.

Behind those sit surveillance, private databases shared between properties, facial recognition, and in some states a statute letting a licensee detain someone suspected of a gaming crime. Counting is not one, and detentions on that basis have produced lawsuits rather than charges.

One thing that does not happen: counters are not placed in Nevada's List of Excluded Persons, which is for cheats and career criminals. The distinction the industry draws in its own regulations is the one it blurs in public.

New Jersey is the exception, and it is narrower than you have heard

In Uston v. Resorts International Hotel (1982), the New Jersey Supreme Court held that a casino could not exclude Ken Uston for counting. The reasoning was jurisdictional rather than moral: the Casino Control Act gave the Commission exclusive authority over the rules and method of play, so a licensee had no leftover common-law right to bar a player over how he played.

What followed matters as much as the holding, because the regulator then approved countermeasures. In Atlantic City a counter generally cannot be barred for counting alone, and can still be shuffled on, flat-bet, and kept out of a shoe in progress. The exception protects your seat, not your edge. It is also a New Jersey ruling about a New Jersey statute, and it does not travel: Nevada has nothing like it, and no other state has adopted the same reasoning.

Devices and teams sit in a different area of law entirely

Nevada makes it a felony to use, or to possess with intent to use, a device that keeps track of the cards played, projects the outcome, or analyses the odds or the strategy at a licensed game (NRS 465.075). Other states have statutes in the same family, but they are not copies of each other: what counts as a device, and whether possession alone is enough, varies, and some jurisdictions have no such provision at all. The Nevada statute is about the device, not the difficulty — it does not care that your phone is only performing arithmetic you could do in your head. Practise on a screen at home; count in your head at the table.

Team play is subtler. A spotter flat-betting a table and signalling a big player is, standing alone, gestures and speech: no device, no altered game. Prosecutions of counting teams have generally turned not on the signalling but on what came with it: false identification, play on someone else's rated account, comp fraud, and structuring cash to stay under the federal currency-reporting threshold. That threshold is set by federal regulation; look it up rather than trusting a figure in an article. Structuring is a federal crime in its own right, whatever the money is.

And when a team reaches across to the dealer's side of the table, a false shuffle or a cooperating dealer, it stops being advantage play and becomes the thing the cheating statutes were written for. Those cases get prosecuted, including federally.

What "illegal" usually turns out to mean

Three confusions do most of the damage.

Banned is not illegal. Being told to leave is a property decision a private business makes in four seconds. No record, no charge, no court.

"Not a crime" is not "the win stands." Phil Ivey was never charged with anything. He also did not keep it. The UK Supreme Court held in 2017 that his edge sorting was cheating within the meaning of the Gambling Act, and a federal court in New Jersey held the same conduct breached the Casino Control Act and ordered repayment to the Borgata. Criminal law, gaming regulation and the contract you enter when you sit down are three separate questions, in two different legal systems in that case, and you can win one while losing the others.

Counting, hole-carding and edge sorting are not one thing. Courts have broadly tolerated playing off a card a dealer exposes through their own error, provided the player did nothing to cause the exposure, and have been far less forgiving when the player engineered it. Those outcomes have varied by state and turned hard on the facts. Counting is the cleanest of the three because it asks nothing of the house.

Two more, briefly. Tribal casinos are sovereign, governed by tribal law and the state compact, so your recourse there is not the recourse you have in a commercially licensed room. And all of this is US law; other countries reach broadly the same answer by different routes, with real variation in the countermeasures a regulator permits and in what a casino may do with a player's winnings after the fact.

The question that matters more than the legal one

Say nothing stops you: no back-off, no trespass notice, a seat for as long as you want it.

Two definitions first, because every figure after them depends on both.

Penetration is how far into the shoe the dealer goes before shuffling, as a share of it. On six decks, 75% penetration means four and a half decks are dealt and one and a half stay behind the cut card, never seen.

True count is the running count divided by the number of decks still to be dealt. The running count is a tally: in Hi-Lo, 2 through 6 are +1, 7 through 9 are 0, tens and aces are −1, added up as the cards fall. The tally on its own is not the number you act on. Say three decks of a six-deck shoe are in the discard tray, so three are left to deal, and your running count is +6. Six divided by three is a true count of +2. That same +6 with five decks still to come is a true count of +1.2 — identical tally, much weaker game. The divisor is an estimate you take off the discard tray by eye, and being half a deck out is enough to change which plays fire. Every count in the rest of this article is a true count.

Now find the good game: six decks, dealer stands on soft 17, double after split, blackjack pays 3:2, 75% penetration. Basic strategy faces a house edge of 0.41% there. Change that one line — dealer hits soft 17, everything else held — and it is 0.63%. Leave soft 17 alone and change the payout instead: 6:5 costs 1.36 points on a six-deck game, so 0.41% becomes 1.77%. Make both changes, H17 and 6:5, and the same table is 1.99%. That is two lines changed, not one. No single rule on that list takes you from 0.41% to 1.99%, and any figure quoted without its rule set is unreadable.

The hit-soft-17 version is worth looking at twice, because the cost and the work are unrelated. H17 takes 0.22 points off you and changes exactly three cells of the six-deck chart: soft 19 against a 6, soft 18 against a 2, and hard 11 against an ace, each of which becomes a double. Three cells. The rule takes the money structurally, not by making you relearn the game.

Counting moves the number by telling you when to bet more. Hi-Lo is worth roughly 0.5 points of edge per true count, with a band of 0.45 to 0.55. On the stand-on-soft-17 game above, a true count of +2 — the +6 with three decks left — puts the player at +0.59%. On the hit-soft-17 version of the same game, +0.37%. The line does not cross +1% until true count +2.8 on the stand game, or +3.3 on the hit game. Those counts are rare: in a six-deck shoe, 14.9% of rounds are dealt at true count +2 or better, and 4.7% at +4 or better.

Which is why the bet spread is the whole mechanism. Count perfectly on that six-deck, S17, DAS, 3:2, 75%-penetration game, play all 22 indices (Illustrious 18 plus Fab 4), and bet the same amount on every hand, and you sit at about −0.2%. Flawless play, still the wrong side of zero. That is the sharp end of the legal question: what is legal is not what is tolerated, and the behaviour that creates the edge — a bet that grows as the true count grows — is the one behaviour the pit is trained to watch for.

Assume it is tolerated anyway. The Wizard of Odds reference ramp — a 1-to-10 spread, 4.5 of six decks dealt, Illustrious 18 plus Fab 4 — gives an advantage of 0.587%, an average bet of 1.68 units, and a standard deviation of 2.27 units per round. One caveat, and it carries through every figure derived from that ramp for the rest of this article: the game they simulated is more liberal than the 0.41% floor described above. Treat the advantage, the bankroll and the hours below as an upper bound on the game you will actually find on a floor. A skilled counter's realistic advantage, across playable rule sets rather than any one of them, is 0.5% to 1.5%.

Over 100 hands, the swing around expectation is 23 to 1. Over an hour it is worse, not better: a six-deck table with four or five players deals about 70 hands an hour, and 70 hands is 27 to 1. Swing grows with the square root of the hands while expectation grows linearly, so the ratio is 230.2 divided by the square root of the rounds — check it: 230.2 over the square root of 100 is 23. It shrinks with volume, and it shrinks slowly. It takes tens of thousands of hands to reach 1 to 1. Not that you might have a bad night. That the noise is more than an order of magnitude larger than the signal in any session a person actually sits through.

The bankroll follows from that, not from ambition. On that ramp, a 5% risk of ruin needs 782.5 minimum bets, which at a $25 table is $19,562 you must be able to lose without changing how you bet. A 1% risk of ruin needs 1,202.9 units.

Then the clock. N₀ — the hand count at which expectation finally equals one standard deviation, the 1-to-1 point above — is 52,960 hands on that ramp, about 757 hours at 70 hands an hour. Standing there the whole way, the chance you are still behind when you get there is about 16%. That is not bad luck. It is the one-sigma tail, and it is what N₀ means.

The games are also thinning. On the Las Vegas Strip, 275 of 855 tables still paid 3:2 as of September 2022 (Vegas Advantage), down from 720 in 2010. Zero single-deck 3:2 games remain in Las Vegas; the last closed at El Cortez on 28 April 2024. Penetration is worth more than most people think: at a fixed 1-to-10 spread, 4 of six decks dealt is 0.368%, 4.5 of six — the 75% defined above — is 0.587%, and 5 of six is 0.837%. Where the cut card sits is worth more than most rule lines, and it is the one thing on the table nobody publishes. The small decisions are just as unsentimental: insurance on six decks, 96 tens among 311 unseen cards, is a 30.87% shot needing 33.33% to break even.

If you want to know what a particular set of rules is worth, compute it rather than reading it off someone's chart. The Cut Card's chart page is free and needs no account: set the rule set and it prints the strategy computed from those rules, with the house edge that belongs to them, unflattering figures included. It runs at your kitchen table, which under NRS 465.075 is where electronic help has to stay.

The short version

Counting is legal in the United States, and almost everywhere you can still be shown the door for it on the spot with nothing to appeal to. Devices are a felony in Nevada. New Jersey protects the seat, not the edge.

The legal question has a clean answer, which is why it is the one everybody asks. The questions without clean answers have arithmetic in them: whether you can hold a running count and convert it through eight hours of noise, whether you can find 4.5 decks dealt out of 6, whether you can watch $19,562 move around without flinching, whether you are still at the table at hour 757.

Most people who buy counting training never play a counted shoe. Nobody has a real number for that, including us, and anyone quoting one is guessing. The shape is not in doubt though, and it was never the law that stopped them.