Baccarat card counting: tiny shifts, rare edges

Baccarat card counting means keeping track of which cards have left the shoe and changing your bet when the cards that remain favour one side. It is possible in principle, because dealt cards are not put back. In practice the main bets hardly move. A count can lower the cost of playing banker and player by a few hundredths of a percent; it cannot make either one a winning bet. The places where counting can matter are a couple of side bets, late in a shoe that is dealt nearly to the end. This page is the deeper part of our baccarat strategy guide, with the numbers behind each claim.

Why baccarat card counting barely moves banker and player

The measure counters use is the effect of removal: how much the expectation of a bet changes when one card of a given rank leaves a full shoe. The Wizard of Odds publishes these values for an eight-deck shoe. Converted to percentages, a few of them look like this:

The strongest card for banker, the 4, improves the bet by about a ten-thousandth of the stake. Against a banker edge of 1.06%, you would need the net effect of roughly ninety such cards, all leaning the same way, to reach break-even. Real shoes do not deplete like that. Low and high cards leave together, and the effects cancel.

A second reason is that baccarat has no decisions after the bet. Both hands draw by a fixed table, so the only lever is which side you back and how much you stake. The full rules are on baccarat rules.

Our own worked illustration shows how stubborn the edge is. Take a full eight-deck shoe and remove all 32 fours, the most banker-friendly rank, and nothing else. An exact calculation of every possible deal gives:

Even this impossible-looking shoe leaves banker at a loss. That is the core of why baccarat card counting disappoints on the main bets.

What the research says about the tie

Edward Thorp and William Walden proved the Fundamental Theorem of Card Counting in 1973. It says the spread of possible player expectations widens as a shoe is used up. They applied it to the tie bet paid at 9 for 1 (8 to 1) and concluded that significant favourable counting strategies for the tie do not exist. In their simulation, a tie advantage one standard deviation above average only showed up once 26 or fewer cards remained, and they noted that shuffling at about that point removes it. The same paper contrasts this with their earlier finding that side bets on a natural 8 and 9 could be counted, bets that had disappeared by 1973. More on the tie itself is on the tie bet page.

What a count saves on banker and player

The Wizard of Odds describes a simple level-one count: +1 for each ace to 4, −1 for each 5 to 8, nothing for 9s and ten-value cards. You bet player when the running count is −4 or lower and banker otherwise. The combined edge drops to 1.01%, against 1.06% for banker every hand. If you also sit out counts from −8 to 0, you play 55.69% of hands at an edge of 0.95%.

Our illustration of what that is worth at $25 a hand over 100 hands dealt:

The last row looks like a big saving, but most of it comes from betting half as often. Per hand actually bet, the count saves about three cents. The choice between banker or player matters about as much as the count does, and it costs no effort.

Side bets: where baccarat card counting can matter

Some side bets depend on a narrow set of cards, so their odds swing far more than banker or player. Two have published counting research. Both belong to EZ Baccarat, and the figures below are for eight decks. Commission-free tables in general are covered on no-commission baccarat.

• Dragon 7 pays 40 to 1 when banker wins with a three-card 7. Off the top of an eight-deck shoe the house edge is 7.61%.
• Panda 8 pays 25 to 1 when player wins with a three-card 8. Off the top the house edge is 10.19%.

Eliot Jacobson's analysis of the Dragon 7, published on the Wizard of Odds, uses a simple count: −1 for each 4, 5, 6 and 7, +2 for each 8 and 9. You bet the Dragon at a true count of +4 or higher. That happens on 9.163% of hands, roughly seven bets per shoe, at an average advantage of 8.032% per bet. Over a shoe that is worth about 0.5967 units, or 0.440 units per hour at 60 hands per hour. Jacobson's own verdict: given the high variance and low return, counting is not a practical threat to the game. Bill Zender, writing for casino operators in 2023, describes the same count and says traditional counting cannot beat the main game.

The Panda 8 count on the Wizard of Odds bets at a true count of 11 or more, which comes up on 4.61% of hands, at 6.34% per bet and 0.238 units per shoe. The page rates the Dragon about 2.5 times as profitable to count.

To see why the Dragon reacts so strongly, here is our calculation for two decks left in the shoe, with some 8s and 9s already gone:

Banker and player barely move. The Dragon goes from a bad bet to a strongly positive one. The catch is frequency: even in the best row the Dragon wins only about 2.8% of the time. At around 2.5% per bet, seven bets in a shoe all lose about five times in six. The positive number is real, and so are long runs of losing shoes.

Shoe penetration: why the cut card decides it

Penetration is how much of the shoe is dealt before the cut card comes out, and it is the single biggest factor in baccarat card counting. Thorp and Walden's theorem explains why it decides everything: the useful moments cluster at the very end of the shoe. The research above shows it in numbers.

1. Jacobson's Dragon 7 figures assume the cut card sits 14 cards from the end. Moving it to one deck from the end roughly halves the possible profit.
2. The Panda 8 count earns 0.238 units per shoe with the cut card between the 14th and 15th card from the end. With 104 cards cut off, it falls to 0.02 units.
3. For the tie, Thorp and Walden found that shuffling with 26 cards left removes the only meaningful advantages.

A deep cut is the condition every counting result depends on. Without it, there is nothing left to count.

Online tables: what the game rules say

Baccarat card counting online depends on one detail: whether the game keeps a shoe at all. Do not assume that an online table reshuffles every hand. Evolution's 2020 announcement of First Person Baccarat, its RNG version, quotes its Chief Product Officer: the game deals from a true eight-deck shoe and does not digitally shuffle after each round, and players can shuffle the shoe and cut the cards themselves. For other RNG tables, the official pages we opened do not say how often the shoe is shuffled. The game's own help panel is the place to check.

Live tables deal from a physical shoe on camera. The studio material we opened does not publish where the cut card goes, so treat the position as unknown until you see it in the game rules or on screen. Pattern sheets on the screen are a different thing: the Big Road records results, not the cards left. Our page on live baccarat covers how those tables run.

The score counts what each casino publishes about its licence, ID checks and withdrawal limits, not how its baccarat lobby looks.

A checklist before you try baccarat card counting

1. Pick the bet first. On banker and player a count saves cents per hand. Of the side bets, only the Dragon 7 and Panda 8 have published counting research.
2. Find the cut card. If it sits a deck or more from the end, most of the gain is gone.
3. Read the game's help panel. Look for the number of decks, the side-bet pay table and how often the shoe is shuffled.
4. Size for the swings. A 40 to 1 bet that wins about once in 40 can lose for many shoes in a row.
5. Set a loss limit before you sit down. A count changes the odds a little and the variance a lot. See responsible gambling.

If you prefer simply to play the lowest-cost bet, the streak and pattern questions are covered on baccarat streaks.

Questions and answers