GuideCounting, worked out
Baccarat Card Counting: Does It Work? The Exact Numbers, Shoe by Shoe
Can you count cards in baccarat? You can keep track of them, and the odds do shift as the shoe runs down. The real question is whether they shift far enough, often enough, to matter. We worked it out exactly for the last 30 hands of 100,000 simulated shoes.
What counting cards would have to do in baccarat
You get one decision at this table, which box gets your stake, and a rule settles everything after it. Counting is an attempt to make that one decision with more information.
Some players look for a way around the fixed cost of the game, and baccarat card counting is the one that sounds most serious. If remembering the cards can work at other tables, it seems reasonable that it should dent the house here as well. The idea is to track which cards have left the shoe and bet only when what remains favours your side.
The drawing rule is fixed and automatic. Player draws on a total of five or less. If Player stands, Banker draws on five or less. If Player draws, Banker follows a printed chart that depends on its own total and on the value of Player's third card: Banker always draws on three, for example, unless Player's third card was an eight. Because of that chart, some cards help Banker when they leave the shoe and others help Player.
For a count to pay, the remaining cards would have to lean so far one way that the edge on Banker or Player drops below zero, and you would have to bet only then. In a shoe of 416 cards, a handful of dealt cards hardly tips anything. You would have to sit through dozens of hands, recording every card, waiting for the shoe to lean far enough. The tables below measure how far it leans, and when.
Why the main bets barely move
The house edge on Banker is 1.06% and on Player 1.24%. Taking one card out of the shoe moves those figures by around a hundredth of a percentage point.
Every card value shifts the bets a little. Aces, twos, threes, fours and the tens and picture cards help Banker when they leave the shoe; fives to nines leaving help Player. The largest single effect is the four, which improves Banker by about 0.012 percentage points per card removed. Even if those effects simply added up, wiping out Banker's 1.06% edge would take about 90 fours out of the shoe, and an eight-deck shoe only holds 32.
That is why the main bets barely move. The effects run in opposite directions, many of them cancel inside the same shoe, and the edge each bet starts from is far larger than any one card can shift. The Tie moves several times as much per card, with sixes and sevens leaving hurting it most, but it starts from 14.36%.
| Card removed | Banker | Player | Tie 8:1 |
|---|---|---|---|
| A | +0.0044 | −0.0045 | +0.0129 |
| 2 | +0.0052 | −0.0054 | −0.0239 |
| 3 | +0.0065 | −0.0067 | −0.0214 |
| 4 | +0.0116 | −0.0120 | −0.0292 |
| 5 | −0.0083 | +0.0084 | −0.0264 |
| 6 | −0.0113 | +0.0113 | −0.1159 |
| 7 | −0.0083 | +0.0082 | −0.1091 |
| 8 | −0.0050 | +0.0053 | +0.0654 |
| 9 | −0.0023 | +0.0025 | +0.0426 |
| 10/J/Q/K | +0.0019 | −0.0018 | +0.0513 |
Own exact calculation: change in each bet's expected return, in percentage points, when one card of that value is taken out of a full eight-deck shoe. Plus means better for the bettor.
Side bets and late-shoe spots
Counting only matters when a bet turns positive, and that happens almost exclusively in the last hands before the cut card.
Before every one of the last 30 hands of each shoe, we worked out the exact odds of Banker, Player, the Tie at 8 to 1 and the Player Pair at 11 to 1 from the cards still in the shoe. Across all 3,000,000 hands, Banker was positive on 0.98% of them and Player on 0.83%. When Banker did turn positive, its average edge for the player was 0.51%; for Player, 0.68%.
In the last five hands of a shoe, with about 22 cards left, Banker was positive 5.2% of the time and Player 4.3%. In hands six to ten from the end they fell to about 0.6%, and further back almost to nothing. A separate check of 6,000 shoes, with every hand worked out exactly, found no positive Banker, Player or Tie spot in any hand before the last 20.
The side bets move more, as system sellers like to point out. The Player Pair turned positive on 7.15% of the last 30 hands, and in the last five hands on 25.7%, because a small pool of cards can easily hold several of one rank. But the Pair starts from a house edge of 10.36% and the Tie from 14.36%, so the shift has to be large just to reach zero. A slight bump in probability does not turn an expensive bet into a good one; most of the time it only makes it slightly less expensive.
| Hands from the end | Cards left, average | Banker | Player | Tie 8:1 | Player Pair 11:1 |
|---|---|---|---|---|---|
| Last 5 | 22 | 5.24% | 4.29% | 5.06% | 25.67% |
| 6 to 10 | 47 | 0.58% | 0.60% | 0.18% | 11.82% |
| 11 to 20 | 84 | 0.03% | 0.03% | 0.00% | 2.61% |
| 21 to 30 | 133 | 0.00% | 0.00% | 0.00% | 0.10% |
Own simulation: 100,000 eight-deck shoes, seed 20261106, run 6 November 2026. Exact odds from the cards left before each of the last 30 hands of every shoe; cut card 14 cards from the end.
What a perfect counter would earn
Suppose you knew every card left in the shoe, worked out the exact odds before each hand and bet £10 only when Banker or Player was positive.
That perfect counter would place a bet on 0.67 hands in every 100 dealt, and only 28% of shoes offered even one such spot. The average edge on those bets was 0.59%. Put together, the expected gain comes to about 4p per 100 hands dealt, or 3.2p per shoe.
For comparison, flat £10 on Banker for the same 100 hands costs about £10.58 on average. Anyone who plays the hands is therefore at a net loss, count or no count: the 4p exists only for a perfect counter who sits out more than 99 hands in every 100, with a perfect memory of all 416 cards and exact arithmetic before each deal. Every hand played in between costs its usual 1.06% on Banker, and a count kept by memory at a live table's pace gets nowhere near perfect.
You would sit at a table for hours, not betting, waiting for one spot near the cut card. The margin is too thin to matter, which is why the odds and strategy page comes back to the same plain choice: Banker, no Tie, no side bets.
Online and live tables: shuffles and cut cards
The biggest barrier to counting is table procedure. A count needs the end of the shoe, and that is the part the table never deals.
Live tables shuffle early, either with a continuous shuffle or with a cut card placed before the end of the shoe. When the cut card comes out, the hand in play is finished, one more hand is dealt, and the shoe is changed. The cards behind the cut card are never dealt, and those are the ones where a count could matter most. All your tracking resets with the new shoe, just as the composition might have been getting interesting.
Our simulation placed the cut card 14 cards from the end. A cut card placed earlier leaves even fewer late-shoe spots than the tables above show, because the hands where Banker and Player turn positive are the last few before it.
A live squeeze table changes none of this; a slowly turned card is the same card as a fast one. The other betting ideas that rely on the shoe, such as following streaks, are covered on the betting systems page. The only choice that makes sense is the simplest one: back Banker, pay the commission and leave the side bets alone.
How the numbers were worked out
The figures come from our own shoe simulator, run on 6 November 2026 with seed 20261106.
We dealt 100,000 eight-deck shoes by the standard rules: shuffled by a seeded random generator so the run can be repeated, first card turned and that many burned, cut card 14 cards from the end, one more hand after it comes out. Before each of the last 30 hands of every shoe, the odds of Banker, Player and Tie were worked out exactly from the cards left, by going through every possible order of the next four to six cards; the Player Pair odds come from the ranks left. A perfect counter bet only when Banker or Player was positive. The removal table uses the same exact method on a full shoe. Limits: one cut-card depth, eight decks, a counter with perfect memory; real play would do worse.
Questions about counting cards in baccarat
Can you count cards in baccarat?
You can keep track of the cards, but it does not turn the game into a winning one. A perfect counter in our simulation, betting £10 only on the rare hands where the odds tipped, expected about 4p per 100 hands; playing all 100 of those hands at £10 on Banker costs about £10.58, so a player who plays the hands stays at a net loss.
Does the cut card stop a count?
It cuts it short. The cards behind the cut card are never dealt, so the late spots a count could use are lost before they arrive. Our simulation placed the cut card 14 cards from the end, and even a perfect count gained almost nothing on the main bets.
How often does the Banker bet turn positive?
Very rarely: on 0.98% of the last 30 hands of a shoe in our simulation, and almost only in the last five.
Does the number of decks matter for counting?
More decks dilute each card's effect. Fewer decks shift the edge a little: Banker costs 1.01% with one deck and 1.06% with eight.
Which baccarat bet moves most with the cards left?
The side bets. The Player Pair turns positive most often late in the shoe, but it starts from a house edge of 10.36%, the Tie from 14.36%.
Is counting worth it online?
No. Live tables change the shoe at the cut card or shuffle continuously, so the late-shoe hands where a count could help are rarely dealt.