How to count cards — and why it actually works
Card counting is not memorizing the deck, and it doesn't require an unusual brain. It is tracking a single number that summarizes how the remaining cards differ from average — and betting more when that difference favors you. This guide builds the whole system from first principles: the probability theory that makes blackjack beatable, the Hi-Lo count that exploits it, and the arithmetic that turns a count into an edge measured in dollars.
Four things to know before you start
Most of what people believe about card counting comes from movies, and most of it is wrong in ways that scare off exactly the people who could learn it. The reality:
It's legal
Counting is arithmetic on cards you can see — no US jurisdiction criminalizes it. Casinos can ask a suspected counter to leave, but playing well is not cheating.
No genius required
You track one number, adding or subtracting 1 as cards appear. The challenge is accuracy at speed for hours — discipline, not intellectual horsepower.
The payoff is real, but modest
A well-played game yields a 0.5–1.5% overall edge. The money comes from betting big at the right moments, over volume — not from winning every session.
It takes 2–3 months
Basic strategy, then the count, then deviations. Most disciplined learners are casino-ready in two to three months of regular practice.
The learning path
Here's the whole journey up front, so you know where each piece of this guide fits. Each layer is worthless without the one below it:
- Basic strategy to perfection first. Deviations are adjustments from a baseline; without the baseline there is nothing to adjust. Start with the basic strategy charts.
- Running count drills. Count down decks until a sub-30-second, zero-error countdown is boring.
- True count conversion. Practice estimating decks remaining and dividing under time pressure — this is where most real-world errors creep in.
- Deviations at the table. Layer in the Illustrious 18 and Fab 4, drilled until the index numbers fire automatically at the right counts.
- Bankroll math before the casino. Run your intended game through an EV analysis. Know your hourly EV, standard deviation, and risk of ruin before a dollar goes on the felt.
The rest of this page explains why the system works andhow each piece operates — starting from the one fact that makes any of it possible.
Why blackjack is beatable at all
Roulette cannot be beaten by any betting system because every spin is independent — the wheel has no memory, so no observation of past spins tells you anything about the next one. Blackjack is different in one structural way: cards are dealt without replacement. Every card that leaves the shoe changes the composition of what remains, and the expected value of the next round is a function of that composition. Past observations carry real information about the future. That single fact is the entire foundation of card counting.
Information alone isn't enough — there must also be an asymmetry to exploit. Blackjack has four, and all of them point the same direction: a shoe rich in tens and aces favors the player; a shoe rich in small cards favors the house.
- Blackjacks pay 3:2, but only to you. A ten-rich shoe produces more naturals for both sides equally — but yours pay 1.5 units while the dealer's takes just 1. More blackjacks means more asymmetric payouts in your favor.
- The dealer must hit stiffs; you don't. The dealer's algorithm is fixed: hit to 17, no exceptions. Holding 12–16 against a ten-dense shoe, the dealer busts more often. You, meanwhile, can simply stand and let them.
- Doubles land harder. When you double on 10 or 11, the card you want is a ten. Ten-dense shoes convert more of your doubles, and you've got twice the money riding on them.
- Insurance becomes a good bet. Insurance pays 2:1 and wins when the dealer's hole card is a ten. It's a sucker bet in a neutral shoe — but in a sufficiently ten-rich one, it flips positive. The count tells you exactly when.
The Hi-Lo system: mechanics
So high cards remaining are good for you, low cards are bad. Hi-Lo tracks that balance with one tag per card:
The math, if you want it: where the tags come from
The effect of each rank on your expected value has been computed exactly. Remove a single card from a single-deck game and the player's EV shifts by approximately:
| Card removed | Change in player EV | Hi-Lo tag |
|---|---|---|
| 2 | +0.38% | +1 |
| 3 | +0.44% | +1 |
| 4 | +0.55% | +1 |
| 5 | +0.69% | +1 |
| 6 | +0.46% | +1 |
| 7 | +0.28% | 0 |
| 8 | 0.00% | 0 |
| 9 | −0.18% | 0 |
| 10, J, Q, K | −0.51% | −1 |
| A | −0.61% | −1 |
Approximate effect-of-removal values, single deck, benchmark rules (per Griffin, The Theory of Blackjack).
Read the table and the design of Hi-Lo becomes obvious. Removing a 5 helps you most (+0.69%) — it's the card the dealer most needs to make a stiff hand good. Removing an ace hurts you most (−0.61%) — it's irreplaceable blackjack material. The 8 is almost perfectly neutral. A counting system is just a linear approximation of this table: assign each rank an integer tag roughly proportional to its effect, sum the tags of every card you see, and the sum estimates the cumulative EV shift of everything removed so far.
Hi-Lo compresses the table into three buckets: +1, 0, −1. That compression costs surprisingly little — Hi-Lo'sbetting correlation (how well its tags correlate with the true EoR vector) is about 0.97. Fancier multi-level systems buy the last few percent of correlation at the price of more arithmetic per card, and arithmetic errors cost far more than the residual imprecision. That trade is why Hi-Lo has been the professional standard since the 1960s.
Note the balance: each deck holds twenty +1 cards (four each of 2–6) and twenty −1 cards (sixteen tens plus four aces), so the tags over a full shoe sum to exactly zero. This makes Hi-Lo abalanced count — a correct count always returns to zero when the last card is dealt, which is also what makes it self-checking in practice drills.
The running count (RC) is the cumulative sum. Start at 0 on the shuffle, and update for every card you see — yours, the dealer's, and every other player's. When low cards leave the shoe the count rises (what remains got richer in tens and aces); when high cards leave it falls. The sign convention makes the number you track point in the direction of your advantage.
Two habits make it fast enough for a live table. First,count in pairs: most hands expose cards two at a time, and pairs frequently cancel (a king and a four is zero — no arithmetic at all). Second, count card groups as shapes, not card-by-card: a busted hand of 10-6-9 is −1+1+0 = 0 at a glance. The standard benchmark is counting down a full deck in under 30 seconds, ending at exactly zero, with no errors. Speed matters less than the zero.
From running count to true count
Here's the subtlety most beginners miss: the running count measures a surplus, but your advantage depends on aconcentration. A running count of +6 means six extra high-card units are somewhere in the undealt cards. If five decks remain, that surplus is diluted across 260 cards — barely noticeable per card dealt. If one deck remains, the same six units are packed into 52 cards, and the next round will be dealt from a meaningfully stacked deck. Same running count, wildly different edge.
The fix is a density normalization:
Decks remaining is estimated from the discard tray (in half-deck increments is plenty of precision). The true count is the number every downstream decision runs on — bet size and strategy deviations both key off TC, not RC. This is also why deeppenetration is the most important property of a beatable game: the fewer decks behind the cut card, the larger the divisor shrinks, and the more often the true count reaches profitable territory. (Penetration is one of several things to scout before you sit — seehow to choose a table.)
From count to edge
The relationship between true count and player edge is remarkably linear over the playable range. The rule of thumb, confirmed by simulation across common rule sets:
| True count | Approx. player edge |
|---|---|
| −1 | ≈ −1.0% |
| 0 | ≈ −0.5% (house edge, basic strategy) |
| +1 | ≈ 0% (near breakeven) |
| +2 | ≈ +0.5% |
| +3 | ≈ +1.0% |
| +4 | ≈ +1.5% |
Each true-count point is worth about half a percent. The shoe crosses into player-favorable territory around TC +1, and a TC of +4 or +5 — which a deeply dealt shoe reaches regularly — has you playing with the kind of edge the house normally enjoys over a basic strategy player.
Betting the count
The count would be trivia if it didn't change your behavior. It changes it in two ways: how much you bet, and (in specific spots) how you play the hand. Betting is where nearly all of the money is.
The logic is unforgiving: when your edge is negative — which is most of the time — every dollar wagered has negative expectation, so you bet the table minimum. When the edge turns positive, you want maximum money in action. The ratio between your biggest and smallest bet is your spread (1–8 and 1–12 are common in shoe games), and your overall edge is the wager-weighted average across counts — which is why a flat bettor with a perfect count barely beats the game at all.
The math, if you want it: optimal bet sizing (Kelly)
How much is "maximum"? The theoretically optimal wager is given by the Kelly criterion — bet the fraction of bankroll that maximizes long-run logarithmic growth:
In practice, professionals bet a fraction of Kelly (half-Kelly is common) because overbetting is punished asymmetrically: bet double Kelly and your expected bankroll growth is zero, with brutal variance. The interaction between spread, bankroll, and risk is exactly what the EV Analyzer computes — expected value per hour, standard deviation, and risk of ruin for any game and spread you give it.
Playing the count: deviations
Basic strategy is optimal for a neutral shoe — but you now know the shoe isn't always neutral. A few dozen decisions sit close enough to the EV borderline that a shifted count flips the correct play: 16 vs 10 becomes a stand at any positive count, insurance becomes profitable at TC +3, and a handful of doubles and splits activate as the count climbs. These index plays — the famousIllustrious 18 and Fab 4 — carry the large majority of all play-variation value, and we've published the complete charts with the math behind them:
Frequently asked questions
Is card counting illegal?
No. Card counting is using publicly available information — the cards you can see — and arithmetic. No jurisdiction in the US criminalizes it. However, casinos are private businesses and can refuse service: if they suspect you of counting they may restrict your bets or ask you to leave. Using a device to count is a different matter and is a crime in most gaming jurisdictions.
Do you need a photographic memory or a genius IQ?
No — this is the most persistent myth about counting. You track a single integer, adding 1 or subtracting 1 as cards appear. The challenge is not intellectual horsepower, it is doing something trivially easy with perfect accuracy, at speed, for hours, in a noisy casino while looking like you are not doing it.
How much of an edge does counting actually give you?
A well-played Hi-Lo game with a reasonable bet spread yields an overall edge of roughly 0.5% to 1.5% depending on the rules, penetration, and spread. That is small — the profit comes from volume and bet sizing, not from winning every session. Expected hourly value is your average bet times your average edge times hands per hour.
Does card counting work against continuous shuffle machines?
No. A CSM returns discards to the shuffler continuously, so the deck composition never departs meaningfully from neutral — there is no information to exploit. Counting requires a shoe that is dealt deep between shuffles. Penetration (how much of the shoe is dealt before reshuffling) is arguably the most important variable in whether a game is beatable.
Why does everyone use Hi-Lo instead of a more accurate system?
More complex systems (multi-level tags, ace side counts) squeeze out slightly more theoretical edge, but the gain is small — Hi-Lo already captures about 97% of the betting information available to a linear system. In practice, errors are the dominant cost, and simpler systems produce fewer errors at speed. Hi-Lo sits at the sweet spot of power versus operability.
How long does it take to learn to count cards?
Basic strategy first — typically one to two weeks of drilling. The running count itself takes days to learn and weeks to make automatic; the standard benchmark is counting down a full deck in under 30 seconds without errors. True count conversion and deviations add another few weeks. Most disciplined learners are casino-ready in two to three months of regular practice.