Level 5 · Lesson 19 of 4 · Advanced Player
Card Counting in Baccarat: It Works, and It Isn't Worth It
The mathematical basis
As cards are removed from the baccarat shoe, the composition shifts. Some compositions favour Banker (more low cards remaining), some favour Player (more high cards remaining), some create situations where specific side bets become advantageous. The shifts are small. The 8-deck shoe provides a large buffer that absorbs most of the card removal effect. But they are real.
Edward Thorp demonstrated this in "The Mathematics of Gambling" in 1984. Stanford Wong refined it into a simpler 5-count system. Both reached the same conclusion: theoretically real, practically useless on the main bet. Eliot Jacobson's work at apheat.net provides the most thorough modern analysis.
The maximum edge from main-bet counting is approximately 0.7% on Banker, under optimal conditions, requiring removal of at least 50 cards from the shoe.
Why the main-bet count isn't worth pursuing
Three reasons.
The frequency problem. The 0.7% edge appears on a small fraction of hands. You must play through unfavourable conditions (paying the 1.06% edge) to reach the occasional favourable count. The net expected return across all hands, including the unfavourable majority, is much worse than the 0.7% headline suggests.
The variance problem. Even with a genuine 0.7% edge, you need roughly 500 hands for the edge to manifest reliably over chance variation. At $100 maximum bet on favourable hands, the expected profit across those 500 hands is approximately $350. The variance means you could be down thousands mid-sequence. The risk-to-reward ratio doesn't work for individual players.
The countermeasure problem. Casinos in Macau and London track systematic bet variation. At the Venetian Macao's high-limit salons, surveillance technology is sophisticated and explicit. Bet size changes correlated with count indicators are noticed quickly. The response: early shuffle, which eliminates the end-of-shoe compositions where favourable counts concentrate. You count accurately for an hour and gain nothing because the shoe is shuffled before the count matures.
Eliot Jacobson's summary: "The expected gain from counting the main baccarat bets is very small, the variance is huge, and the bankroll requirement is prohibitive."
Side bet counting: where it actually works
Several side bets change value rapidly as the shoe composition shifts.
The Dragon 7 side bet (in EZ Baccarat, paying 40 to 1 on Banker winning with a three-card 7) is sensitive to the proportion of specific card ranks remaining. When the shoe is rich in certain cards, the Dragon 7's house edge inverts into a player edge. Eliot Jacobson's published research documents this precisely: using a counting system tracking specific ranks, player edges of 1.5% to 9% on Dragon 7 are achievable when the count triggers. The trigger count occurs in a meaningful proportion of shoes (roughly 30% of shoes produce at least one favourable Dragon 7 situation by the final deck).
The Panda 8 side bet (25 to 1 when Player wins with a three-card 8) is similarly susceptible. Some pair bets respond to shoe composition shifts as well.
The high payouts of these side bets (40 to 1, 25 to 1) amplify the value of small probability shifts. This is the same principle that makes them terrible bets during uncounted play and interesting targets during counted play.
The casino response
Casinos in Macau and London are aware of side bet counting.
Early shuffle: shoes are reshuffled when a high proportion of cards remain, eliminating end-of-shoe effects. Dragon 7 maximum bets are often capped at $25 to $100 at most US casinos, limiting absolute profit even when the count triggers.
The Phil Ivey case is instructive, though it concerned edge-sorting rather than counting. Ivey won approximately £7.7 million playing Punto Banco at Crockfords in Mayfair in 2012 by exploiting asymmetries in the card backs. The UK Supreme Court's ruling in Ivey v Genting Casinos [2017] UKSC 67 established that this constituted dishonest conduct under civil law, meaning the casino was entitled to withhold payment. The ruling also clarified the legal principle that casinos have the right to refuse service to identified advantage players and to alter game conditions. Side bet counters operate in this legal context: technically not cheating, but subject to the casino's right to counter.
At the Venetian Macao, surveillance is deployed systematically rather than reactively. High-limit rooms track bet-sizing patterns alongside card removal sequences for any player running through multiple shoes. The mathematics of what a counter's bet pattern looks like is known to the surveillance teams. The window between a favourable Dragon 7 count appearing and the surveillance team responding is measured in minutes. In practice, systematic side bet counting at Macau's high-limit rooms is not viable as a long-term strategy, even if it is briefly correct.
In London, the legal framework established by Ivey v Genting means any casino approached by a systematically advantaged player at a side bet can decline to deal, decline to pay, or ban the player outright. The legal right is confirmed. The practical countermeasure is immediate.
Key numbers
| Method | Maximum player edge | Practical status |
|---|---|---|
| Main bet counting (Thorp/Wong) | ~0.7% on Banker | Not practical: frequency, variance, countermeasures |
| Dragon 7 counting (Jacobson) | 1.5% to 9% | Theoretically viable, heavily countered |
| Panda 8 counting | Similar to Dragon 7 | Theoretically viable, heavily countered |
What edge-sorting and counting have in common: both require the casino's cooperation to set up conditions the casino would not freely offer. Ivey needed the dealer to rotate cards. A counter needs the casino to deal deep enough into the shoe. Both require something the house controls. Both are vulnerable to the house simply choosing not to provide it.
Sources: Eliot Jacobson on baccarat advantage play, UK Supreme Court, Ivey v Genting [2017] UKSC 67, Venetian Macao casino.
Welcome to the lesson on card counting in baccarat.
I'm Annabel. The title of this lesson is honest: card counting in baccarat works, and it isn't worth it. I'll explain both halves of that statement precisely, because the nuance matters.
First, the mathematics.
As cards are removed from the baccarat shoe, the composition shifts. Some compositions favour Banker. Some favour Player. Some create situations where specific side bets become advantageous. The eight-deck shoe provides a large buffer that absorbs most of these shifts. But they are real. Edward Thorp demonstrated this in The Mathematics of Gambling in nineteen eighty-four. Stanford Wong refined it subsequently. Eliot Jacobson has published the most thorough modern analysis at apheat.net.
The best-documented main-bet counting system yields at most approximately zero point seven percent player edge on the Banker bet. This edge appears after fifty or more cards have been removed from the shoe, under a specific shoe composition, using a system that tracks all thirteen card ranks. It doesn't appear often. It doesn't stay long.
Why isn't it worth pursuing? Three reasons.
The frequency problem. The zero point seven percent edge appears on a small fraction of hands. You must play through unfavourable conditions, paying the one point zero six percent house edge, to reach the occasional favourable count. The net expected return across all hands is much worse than zero point seven percent.
The variance problem. Even with a genuine zero point seven percent edge, you need roughly five hundred hands for the edge to manifest reliably over chance. At one hundred dollars maximum bet on favourable hands, the expected profit across those five hundred hands is approximately three hundred and fifty dollars. The variance means you could be down thousands mid-sequence. The risk-to-reward ratio is not viable for individual players.
The countermeasure problem. Casinos in Macau and London track systematic bet variation. At the Venetian Macao's high-limit salons, surveillance is sophisticated. Bet size changes correlated with count indicators are noticed. The response is early shuffle, which eliminates the end-of-shoe compositions where favourable counts concentrate. You count accurately for an hour and gain nothing.
Eliot Jacobson's summary of main-bet counting: the expected gain is very small, the variance is enormous, and the bankroll requirement is prohibitive. He is correct.
Now, side bets. This is where counting becomes genuinely interesting.
Several side bets change value rapidly as the shoe composition shifts. The Dragon Seven side bet, which pays forty to one when Banker wins with a three-card seven, is sensitive to specific card rank proportions. When the shoe is rich in certain cards, the Dragon Seven's house edge inverts into a player edge.
Jacobson's published research documents player edges of one point five to nine percent on Dragon Seven when the count triggers. The trigger condition occurs in a meaningful proportion of shoes: roughly thirty percent produce at least one favourable Dragon Seven situation by the final deck.
Panda Eight, which pays twenty-five to one when Player wins with a three-card eight, is similarly susceptible. Some pair bets respond to shoe composition shifts as well.
The high payouts amplify the value of small probability shifts. This is why Dragon Seven and Panda Eight are terrible bets during uncounted play and interesting targets during counted play.
The practical obstacles for side bet counting are severe. Dragon Seven maximum bets are typically capped at twenty-five to one hundred US dollars at most tables. Casinos have shortened shoe penetration specifically to counter Dragon Seven counting. And the Phil Ivey ruling at the UK Supreme Court in two thousand and seventeen is clear: casinos have the right to refuse service to identified advantage players and to alter game conditions. Even if you count Dragon Seven correctly, the casino can and will respond.
The honest summary: baccarat is not a beatable game through counting in the way blackjack can be. The main-bet count is mathematically real and practically useless. The side bet count is more viable in theory and heavily countered in practice.
Knowing this is useful not as a strategy but as a way of understanding the game's limits. If a product or system claims to give you a repeatable edge in baccarat through "reading the shoe," the mathematics in this lesson tells you precisely why that claim doesn't hold.
Count the knowledge. Skip the count.