Edge per starting bet
divides the average loss by the bet you place before the cards are dealt. It is the figure most people mean by "house edge", and it is the one we lead with.
Every house edge and strategy chart on this site is calculated, not copied. We take a game's written rules, deal out every possible combination of cards, and work out what a player earns on average with the best chart for those rules. This page explains how that works, what the figures assume, where they are approximate and how we check them.
A rules document
The rules of play come from a regulator text, never from a rule of thumb.
A stored rule set
Decks, soft 17, doubling, payouts and the odd rule become settings in one rule set.
An exact calculation
Every starting hand, every dealer finish and every decision is worked out card by card.
Played through the table
The same chart is played through the game code behind the play-money table.
A figure with its assumptions
It reaches the page with the rules it assumes, or it is not published.
The house edge is the average amount the player loses per unit bet over a very long run, written as a percentage.
A house edge says how many units are lost, on average, for every hundred units bet. It says nothing about any single session.
We do not use rules of thumb such as "each rule is worth about this much". Each game's edge is computed from that game's own rules. The calculation follows the cards. It starts from a full shoe, works out the chance of every possible starting hand, then the chance of every way the dealer's hand can finish, and then the result of each decision the player can make. Removing a card from the shoe changes the odds of the next one, and the calculation keeps track of that.
The rules are the input. Number of decks, whether the dealer hits soft 17, whether doubling after a split is allowed, how a natural pays, what a dealer 22 does: each is a stored setting taken from a rules document (see Sources). Change one setting and the whole calculation is run again. That is why the figures on a variant page move when you change a rule, and why a page can show one game under several rule sets.
Two ways of stating the edge appear on the site, and they answer different questions.
divides the average loss by the bet you place before the cards are dealt. It is the figure most people mean by "house edge", and it is the one we lead with.
divides the same average loss by everything you put at risk from your own money in the round. Doubling adds a stake. Splitting adds a stake for each new hand. A game that lets you double, split and redouble often will have more money on the table than the opening bet, so its edge per amount wagered is lower than its edge per starting bet.
In games that give some stakes free, such as free doubles and free splits, the free stakes belong to the house. They are not counted as the player's wager. You can see both figures for Free Bet Blackjack, where the gap between them is small. Both figures use the same average loss. Only the divisor changes.
Our figures assume every round is dealt from a freshly shuffled shoe that has not been touched.
No cards from earlier rounds are removed, and no counting is involved. This is the standard way to state a house edge, and it makes figures for different games directly comparable.
At a real table the dealer deals part of the shoe before shuffling, and the cards already seen do change the odds a little. Over a long run, without counting, that effect averages out. Our play-money tables follow the same idea in a simple form: the shoe is shuffled and dealt, and the rules in force are the ones printed beside the table.
Two points follow from this. The figures describe the rules, not a particular casino's shuffle. And we make no claim about card counting in any game.
Some pages show two edges for the same game.
The usual reason is a rule that regulators or casinos set differently. Whether the dealer stands on soft 17 or hits it is the clearest case. Spanish 21 is played both ways under different rule texts, so its page states the figure for each, and they differ.
A second reason is a rule whose wording is unclear or contested. European Blackjack is the example: in some descriptions a dealer blackjack takes every bet, including doubled and split ones, and in others only the original bet. We have not found an official text that settles it, so the page shows the result for each reading and says which is which.
We never show a single number where the sources give a range. A range stated honestly is more useful than one exact-looking figure that only holds for one set of rules.
To show how a variant differs from ordinary blackjack, we need a fixed game to compare against.
Ours is labelled "comparison rules" on every page that uses it: 6 decks, the dealer hits soft 17, doubling after a split is allowed, and a natural pays 3 to 2. Each of those rules is cited separately to regulator texts.
This game is not claimed to be what you will find at any particular casino. Real tables vary, and no single set of rules stands for all of them. The comparison game exists so that every variant is measured against the same yardstick. When a page says a game is better or worse than the comparison rules, this is the game it means.
For each game we calculate the best action for every hand you can hold against every dealer up card: hit, stand, double, split and, where the rules allow it, surrender or rescue.
The chart we publish is based on your hand total and the dealer's card, which is the form a player can actually memorise and use. The house edge shown is the edge you get by following that chart.
Calculating the best play for each exact combination of cards gives a slightly better result, and we work that out as well. It is informational. It is not the figure we print, because nobody plays from a table that long.
Some games need extra detail. In Spanish 21, for example, the best play on a hand can depend on how many cards it holds, so the chart has more than one layer.
Each chart is a grid, and each cell is one decision. Next to a variant's chart we highlight the cells that differ from the same grid for the comparison game. Those are the decisions you need to relearn for that variant. If you already know the usual play, the highlighted cells are the only part you need to study.
We would rather state where the figures are weak than hide it.
Hands that come from a split are the hardest part to calculate exactly. We use a close approximation: each split hand is played from the remaining cards, with the cards of the other split hands ignored. We measured the error against exact enumeration on sampled split decisions. For the regular blackjack family it is below 0.001 percentage points of house edge. For Spanish 21 and Free Bet Blackjack, the largest error on a single sampled split decision was about 0.0001 of a bet. We have not measured exact error for hands split three or four times. The play-through checks below do cover those cases.
Some games pay fixed cash awards for special hands. Those awards are not part of any edge we show, because they depend on how a casino sets them. Where a game has them, the page says so.
Some games are labelled "modelled reference". That means we calculated a figure, but one or more rules could not be tied to a single official rules document. Either the sources disagree, or no official text covers a rule the calculation needs. The label tells you the figure is a fair reference for the rules described and not a claim about any one table.
Three games have a rules description but no house edge. Blackjack Switch needs a calculation for swapping cards between two hands. Double Exposure Blackjack needs one for a dealer with both cards face up. Australian Pontoon needs one for the dealer's one-card start and the pontoon payouts. We have not built those, and we do not quote a figure from elsewhere instead. British Pontoon has no regulator text at all, so it is described and labelled unverified.
Three kinds of checks run whenever a figure could change.
Probabilities must add up to one, to a tiny fixed rounding margin. Rules that should give identical answers must do so, such as the hole-card check against the no-check version with the original-bet-only rule. Changing a rule in the player's favour must not make the edge worse.
Our chart is played through the same game code that runs the play-money table, using the same shuffling, dealing and settlement. One test played 100 million hands for each of two games. Smaller runs of one to ten million hands cover the other configurations, and each Spanish 21 and Free Bet Blackjack variant also gets a run of 2 million hands. The measured result must land within 4 standard errors of the calculated edge. That limit is fixed before the run and never widened.
Where another source publishes a figure for rules we can match, we compare. Our allowed difference is 0.05 percentage points, also fixed in advance.
If a comparison fails, we do not loosen the limit. We write down why it differs, and we do not publish a figure that depends on the disagreement until we understand it. A small number of published figures do differ from ours. Each has a written note, and in a couple of cases the cause is still unknown. The play-through checks confirm that our code pays what our calculation says, so those gaps most likely come from how the other figures were produced. They do not prove the other figure wrong.
Questions
Calculated. We enumerate the cards and work out the exact average result for the rules. Simulation is used only as a check: we play our chart through the game code for millions of hands and confirm the result matches the calculation within a fixed limit.
Tables differ in decks, soft 17, doubling, splitting, surrender and payouts, and each rule shifts the edge. Our figure is for the rules listed beside it. If your table prints different rules, the printed layout decides and our figure will not match exactly.
It means we calculated the figure for the rules as we describe them, but at least one rule could not be tied to one official document. The figure is a sound reference for that rule set. It is not a statement about how any specific casino runs the game.
We only publish a figure we have calculated and checked. Blackjack Switch, Double Exposure Blackjack and Australian Pontoon each need a calculation we have not built yet. British Pontoon has no official rules text. Quoting a figure from another source would break our own rule.
No. Every figure assumes a freshly shuffled shoe and no knowledge of cards already dealt. Counting is not modelled in any game on this site, and we make no claims about it. The glossary explains the terms used on these pages.