Blackjack House Edge and RTP: The Math, Rule by Rule
A house edge quoted without its rule set is a decoration, not a number. Here is what each switch is worth, and how to add them up for the table in front of you.
The short answer
- One number, stated twice: house edge is the expected loss as a percentage of the original wager, and return to player is 100 minus the house edge. A 0.50 percent edge is a 99.50 percent RTP.
- The convention that causes the arguments: blackjack figures are quoted against the initial bet, not against total action. Under Atlantic City rules that is 0.43 percent against the bet and 0.38 percent against the money actually wagered.
- The one rule that dominates: a natural paying 6 to 5 instead of 3 to 2 costs 1.39 percentage points, more than six times what the soft 17 rule costs.
- What it actually costs you: average opening bet times house edge times hands per hour. The edge is the smallest of the three.
House Edge and RTP Are Two Views of One Number
The same shoe game can run anywhere from about 0.3 percent to more than 2 percent depending on six or seven switches set before you ever see a card.
At legal online blackjack sites those switches are printed in the game’s help panel where almost nobody reads them. This page gives the arithmetic: what each rule is worth, how to add the pieces up for any table you sit at, what your own mistakes cost on top, and what the resulting figure actually means for an hour of play. Every figure below is stated with the assumptions it depends on, because in blackjack the assumptions are most of the answer.
House edge is the expected loss expressed as a percentage of the original wager. Return to player is the mirror image: 100 minus the house edge. A game with a 0.50 percent edge has a 99.50 percent RTP. Neither term describes what happens in a session. Both describe the average of an enormous number of hands, computed by working through every possible sequence of cards and weighting each outcome by its probability.
The wrinkle that trips up most comparisons is the denominator. Blackjack figures are conventionally quoted against the initial bet, not against everything you push forward during the hand. That convention is a problem unique to games where the wager can grow: you double a 10 against a six, you split eights twice, and suddenly one initial unit has become three units of exposure. The alternative measure, the element of risk, divides the same expected loss by the total money actually wagered.
The gap is real and it is quantifiable. Divide one figure by the other and you learn something rarely stated outright: about 13 percent of a basic strategy player’s money goes down after the deal, in doubles and splits.
Why two writers quote different numbers for the same game
This is the single biggest reason published blackjack numbers disagree with each other. One quotes the edge against the initial bet, another quotes it against total action, and the two figures for an identical game differ by more than a tenth of a point. Everything on this page uses the initial-bet convention unless it says otherwise, which is the standard used by game mathematicians and by the rule-effect table below.
The Baseline, Stated in Full
Rule effects have to be measured against something. Every figure further down this page is a difference from the game defined here.
The benchmark rule set
- Eight decks
- Dealer stands on all 17s, including soft 17
- Double on any first two cards
- Double after splitting allowed
- Split to four hands, no resplitting aces, one card only to split aces
- No surrender
- Dealer peeks for blackjack with a 10 or ace showing
- Blackjack pays 3 to 2
- Player uses total-dependent basic strategy
That game runs an expected loss of about 0.50 percent of the opening bet, or 99.50 percent RTP. The derivation is worth showing, because it is the same method you will use on every table: Atlantic City rules are this exact game plus late surrender and carry a published edge of 0.43 percent. Late surrender is worth roughly 0.07 percentage points. Remove it and the edge climbs to about 0.50 percent.
Published baselines disagree, and the reasons are legitimate
Reputable sources put the same nominal rule set anywhere from about 0.41 percent to 0.50 percent. The differences come from whether the simulation deals from a cut-card shoe or reshuffles every hand, whether the player uses total-dependent or composition-dependent strategy, and how splits are resolved in the model. Expect published absolute figures for one rule set to disagree by up to about 0.05 percentage points. The rule-by-rule differences are far more stable than the absolute baselines, which is why the method below works even when two sources cannot agree on a starting number.
Deck count on its own
One published table isolates deck count cleanly. It assumes a different rule set from the benchmark above, so read it for shape rather than for absolute values: dealer hits soft 17, double on any two cards, double after split, resplitting any pair including aces up to four hands, no surrender, 3 to 2, dealer peeks, and a continuous shuffler.
| Decks | House edge | RTP |
|---|---|---|
| 1 | 0.014 percent | 99.99 percent |
| 2 | 0.341 percent | 99.66 percent |
| 4 | 0.499 percent | 99.50 percent |
| 6 | 0.551 percent | 99.45 percent |
| 8 | 0.577 percent | 99.42 percent |
Notice how the curve flattens. Going from eight decks to six buys about 0.03 points. Going from two decks to one buys 10 times that. Deck count matters enormously at the bottom of the range and barely at all at the top, which is why an eight-deck online table is not meaningfully worse than a six-deck one, and why a single-deck game that pays 6 to 5 is a trap rather than a bargain.
Where the money really leaves the table
The rules are only one of the levers, and the neighboring guides in this section cover the rest of them.
What Each Rule Is Worth
All figures are in percentage points against the benchmark above. A negative number means the edge falls and the rule favors you.
| Rule change from the benchmark | Effect on house edge | Who it favors |
|---|---|---|
| Single deck instead of eight | -0.48 | Player |
| Two decks instead of eight | -0.19 | Player |
| Four decks instead of eight | -0.06 | Player |
| Five decks instead of eight | -0.03 | Player |
| Six decks instead of eight | -0.02 | Player |
| Dealer hits soft 17 | +0.22 | House |
| Double after split not allowed | +0.14 | House |
| Doubling restricted to 9 through 11 | +0.09 | House |
| Doubling restricted to 10 and 11 | +0.18 | House |
| No doubling at all | +1.48 | House |
| Resplitting aces allowed | -0.08 | Player |
| Drawing to split aces allowed | -0.19 | Player |
| No resplitting of any pair | +0.10 | House |
| No splitting at all | +0.57 | House |
| Late surrender against a 10 | -0.07 | Player |
| Late surrender against an ace | 0.00 | Neither |
| Early surrender against a 10 | -0.24 | Player |
| Early surrender against an ace | -0.39 | Player |
| European no hole card, full rule | +0.11 | House |
| No peek on a 10 up card only | +0.10 | House |
| No peek on an ace up card only | +0.01 | House |
| Blackjack pays 6 to 5 | +1.39 | House |
| Blackjack pays even money | +2.27 | House |
Four observations the table repays
The payout on a natural dwarfs everything
Moving from 3 to 2 down to 6 to 5 costs more than six times what the soft 17 rule costs and roughly triples a typical shoe game’s edge on its own, which is why the 3 to 2 versus 6 to 5 comparison deserves its own page and its own five-second check before you sit.
Surrender is worth less than its reputation
The entire value of late surrender is concentrated in hard 15 and hard 16 against a dealer 10. Surrendering against an ace is worth essentially nothing in a peek game, because the dealer’s blackjacks have already been removed from the equation.
No hole card is not a fairness question
In a peek game you lose only your original bet when the dealer turns over a natural. Without a hole card, the dealer collects your doubles and splits too. That is where the 0.11 points comes from, split roughly 0.08 on doubling and 0.03 on splitting.
These effects are additive approximations
Rules interact slightly, so stacking half a dozen of them can be off by a hundredth or two of a point. That is smaller than the disagreement between published baselines, so it is not worth worrying about.
Building the Number for Any Table
Start at 0.50 percent, apply the differences between the benchmark and the table in front of you, and read off the total.
The method is mechanical. Four real-world examples follow, and two of them are worth setting side by side.
A common online eight-deck game. Eight decks, stands on soft 17, double any two, double after split, no surrender, 3 to 2. Nothing differs from the benchmark, so the edge is 0.50 percent and the RTP is 99.50 percent. This is the shape of most standard RNG blackjack in regulated states, and of a good many live tables streamed to players in New Jersey and Pennsylvania.
A double-deck game with restricted doubling. Two decks (-0.19), dealer hits soft 17 (+0.22), doubling restricted to 10 and 11 (+0.18), double after split allowed, 3 to 2. Total: 0.50 – 0.19 + 0.22 + 0.18 = 0.71 percent. The fewer decks look attractive on the sign outside. The doubling restriction quietly takes the gain back and more.
A liberal six-deck game: 0.33 percent
- Six decks, worth -0.02
- Dealer stands on soft 17
- Double on any two cards, double after split allowed
- Late surrender, worth -0.07
- Resplitting aces, worth -0.08
- Total: 0.50 – 0.02 – 0.07 – 0.08 = 0.33 percent
- Published simulations of that exact game put it at 0.28 percent
A six-deck 6 to 5 game: 2.09 percent
- Six decks, worth -0.02
- Dealer hits soft 17, worth +0.22
- Double after split allowed, no surrender
- Blackjack pays 6 to 5, worth +1.39
- Total: 0.50 – 0.02 + 0.22 + 1.39 = 2.09 percent, an RTP of 97.91 percent
- The two rule changes past the payout are almost noise next to the payout itself
That second game costs you more than four times what the first example does. And the liberal game is a fair check on the method: the five-hundredths of a point of daylight between 0.33 and the simulated 0.28 is the cut-card and strategy-model slop described earlier. Close enough to act on, and a fair illustration of the precision the method actually delivers: trust the first two decimal places, not the third.
What Imperfect Play Costs
Every figure above assumes correct basic strategy on every hand, and that assumption does a great deal of work.
Pages that quote 0.5 percent without flagging it are describing a game almost nobody plays. Regulators are candid about this in a neighboring context: New Jersey’s standard for skill-dependent games, at N.J.A.C. 13:69E-1.28A(g), requires the theoretical RTP to be calculated on the assumption that the player uses optimal strategy, and requires the operator either to disclose that strategy or to supply enough information for the player to work it out. The published number is a ceiling, not a forecast.
Two of the most common departures can be priced exactly, because the rule-effect table already contains them. A player who never doubles is playing a game 1.48 points worse than the quoted figure. A player who never splits gives up another 0.57. Timidity, in other words, is the expensive error, and it is expensive in a way that never announces itself, because the hands you fail to double are hands you often win anyway.
Insurance, computed at the table
Insurance is the error that can be computed from first principles. In a six-deck game the shoe holds 312 cards, 96 of them 10s. With the dealer’s ace exposed, 311 cards remain unseen and 96 of them make a blackjack, so the dealer completes a natural 30.87 percent of the time. Insurance pays 2 to 1, so the expected return per unit is (2 x 0.3087) – 0.6913 = -0.074. That is a 7.4 percent house edge, roughly 15 times the base game’s.
Now scale it. A dealer shows an ace on about one hand in 13, or 7.7 percent. Insurance costs half your bet. A player who insures every time gives up 0.077 x 0.5 x 0.074 = 0.0029 units per hand, which is 0.29 percentage points added to the game’s edge. Always insuring turns a 0.50 percent game into roughly a 0.79 percent game, a 58 percent increase in your expected loss for a habit most players think of as cautious.
Taking even money on your own blackjack is the same bet wearing a friendlier name. Holding an ace and a 10 against a dealer ace, 309 cards are unseen and 95 of them are 10s, so the dealer has a natural 30.74 percent of the time. Declining even money returns 0.6926 x 1.5 = 1.039 units on average. Even money returns exactly 1.000. The habit costs 3.9 percent of your wager every time you accept it, though because the situation is rare it does far less aggregate damage than blanket insurance.
| Habit at the benchmark game | Effect on the edge |
|---|---|
| Never doubling | +1.48 points |
| Never splitting | +0.57 points |
| Insuring every dealer ace | +0.29 points |
| All three, stacked on a 0.50 percent game | About 2.84 percent in total |
Add the pieces and you get a defensible upper bracket. A player at the benchmark game who never doubles, never splits and always insures is playing 0.50 + 1.48 + 0.57 + 0.29 = about 2.84 percent, five and a half times the number on the marketing page, before a single side bet. Most players are somewhere between the two extremes. There is no audited measurement of what the average online blackjack player actually gives up, and anyone quoting one precisely is guessing, but the direction is not in doubt and the magnitudes above bound it.
The errors worth fixing first
Ranked by how much money passes through them:
- Refusing to double soft hands and 9 against a small dealer card
- Refusing to split eights and aces
- Hitting hard 12 through 16 against a dealer 2, 3, 4, 5 or 6 out of fear of the small card
- Standing on soft 17 or soft 18 instead of hitting or doubling
- Splitting 10s
- Taking insurance
Why the worst habit is the most durable one
Correct play on stiff hands against a weak dealer card feels wrong precisely because you bust so visibly, which is why it survives so long.
Side Bets Are Where the Hold Comes From
The base game is one of the cheapest propositions in any casino. The circle next to it usually is not.
Side bet house edges run from a few percent to over 25 percent, and the figure depends entirely on the pay table and the number of decks, so a name alone tells you nothing. Published analyses give these ranges:
| Bet | House edge range | What drives it |
|---|---|---|
| Insurance, six decks | About 7.4 percent | Deck count only; the pay is fixed at 2 to 1 |
| Perfect Pairs | About 2.2 percent to 26 percent | Pay table and decks; fewer decks are much worse here |
| Buster Blackjack | About 6.2 percent to 8.8 percent | Pay table, decks and the soft 17 rule |
| Lucky Ladies | About 5.2 percent to 30 percent | Pay table above all; the spread across versions is enormous |
Read those ranges as a warning about naming conventions rather than as a shopping list. Two games labeled Lucky Ladies can differ by 25 percentage points. The pay table governs, and on an online table the pay table is in the help panel, one tap away, which is a genuine advantage online play has over a crowded pit where the felt is half obscured.
Perfect Pairs runs the other way
It gets cheaper as decks are added, because pair frequency rises with more decks, exactly the opposite of the base game. Do not carry base-game instincts into the side circle.
A side bet can trigger paperwork the base game never does
Ordinary blackjack almost never produces a Form W-2G, because table games are absent from the reporting categories and the base game cannot reach the catch-all rule that requires a payout of at least 300 times the wager. A side bet paying 300 to 1 or better can and does reach it, complete with withholding, which is covered in more detail on our page about taxes on blackjack winnings.
The costs that sit outside the house edge
Side bets, bonus math and tax paperwork all attach to money the base-game figure never counted.
Expected Loss per Hour
House edge is a rate. What it costs you is a function of three things, and only one of them is the edge.
The whole formula
Expected hourly loss = average opening bet x house edge x hands per hour.
Hand rates at a live table are well documented. Figures published in Stanford Wong’s Professional Blackjack give the following, and a typical seated player at a busy table is therefore in the 56 to 90 range. The widely repeated “60 to 80 hands an hour” is a fair summary of the seat most people occupy.
| Players at the table | Hands per hour |
|---|---|
| Heads-up with the dealer | 248 |
| Two players | 158 |
| Three players | 116 |
| Four players | 91 |
| Five players | 76 |
| Six players | 64 |
| Full table of seven | 56 |
Online is a different animal. A solo RNG table deals as fast as you can tap, with no shuffle, no chip handling, no players to wait for and no dealer conversation. Two hundred hands an hour is unremarkable and 400 is achievable. Games streamed from a studio sit in between: the fixed betting window and the dealer’s pace put live dealer blackjack below solo RNG speed, often below a live pit table.
The table below applies the formula at a 0.50 percent edge, in dollars per hour.
| Opening bet | 60 hands | 80 hands | 200 hands | 400 hands |
|---|---|---|---|---|
| $5 | $1.50 | $2 | $5 | $10 |
| $10 | $3 | $4 | $10 | $20 |
| $25 | $7.50 | $10 | $25 | $50 |
| $50 | $15 | $20 | $50 | $100 |
| $100 | $30 | $40 | $100 | $200 |
Speed is a multiplier on the edge, and a larger one than any rule on the table
A $25 player at 400 hands an hour on a perfect 0.50 percent game loses $50 an hour in expectation. The same player at a live table at 70 hands loses $8.75. Identical edge, identical strategy, nearly six times the cost. Moving from 3 to 2 to 6 to 5 quadruples your expected loss; quadrupling your hand rate does the same thing and does it invisibly, because nothing on screen changes.
Stack the two and the arithmetic turns unpleasant fast. A $25 player at 400 hands an hour on a 6 to 5 game with hit-soft-17, playing it imperfectly at an effective 2.5 percent, is facing 25 x 0.025 x 400 = $250 an hour of expected loss on a game advertised as the best odds in the building.
Time yourself
Count hands for five minutes and multiply by 12, and you will know your real rate rather than a guess.
Use the tooling you are owed
Regulated sites are required to give it to you, and deposit and loss limits are the only control that acts on all three variables at once.
Run bonuses through the same arithmetic
A wagering requirement is denominated in hands, not in hours, which is the point our guide to bonus terms and wagering requirements keeps returning to.
Pace, format and the controls that cover both
Hand rate is set by the format you choose and by the limits you set before you start.
How RTP Is Certified and Reported
A certified RTP is a statement about the long run and about the software, not a promise about your afternoon.
It is a mathematical property of the game as coded, verified by an independent laboratory against the submitted paytable and rules. The verification process itself is covered in our explanation of RNG certification and testing.
Regulated states also require that theory be checked against practice. New Jersey’s rule at N.J.A.C. 13:69O-1.9(j) requires an internet gaming system to generate a Performance Report comparing the theoretical return to patron against the actual return for every game offered, together with the total number of rounds played, reviewed monthly by the licensee and covering the period from the first day the game went live. That last clause matters more than it looks: the comparison window is cumulative from launch, so a game cannot hide a persistent shortfall behind a convenient month.
New Jersey sets that slot floor at N.J.A.C. 13:69E-1.28A(a). Blackjack at 99.5 percent is not in the same category of product, and the gap explains why Pennsylvania taxes online table games at 16 percent while taxing online slots at 54 percent.
Public revenue reporting is a different thing again, and it is routinely misread. Regulators publish what operators won, not what games returned. The New Jersey Division of Gaming Enforcement publishes monthly internet gross revenue, and the Pennsylvania Gaming Control Board publishes interactive gaming revenue, but neither figure is an RTP.
A published win percentage is not a house edge
The Nevada Gaming Control Board makes the distinction explicit in the introduction to its monthly revenue report: the win percentage it publishes for slot devices is win divided by the total amount played by patrons, while the win percentage for table games is a ratio adjusted for the effects of credit play. Table game win is measured against the drop, meaning money exchanged for chips, not against total wagered. A published table game win percentage in the teens is therefore not a house edge and cannot be compared to one, because the same dollar is bet, won, and bet again many times before it leaves. Our page on reading published payout reports works through the distinction with examples.
Who checks the number, and who publishes it
Certification, monthly performance reporting and public revenue tables are three different things.
Why a Losing Run Proves Nothing
The volatility of a single hand is more than 200 times the edge. Everything people believe they can see in a session sits inside that ratio.
Blackjack’s per-hand standard deviation under basic strategy is about 1.14 units of the opening bet at liberal six-deck rules, against an expected loss of 0.005 units. Standard deviation grows with the square root of the number of hands while expected loss grows in a straight line, which is why the edge takes so long to surface. At $25 a hand on a 0.50 percent game:
| Hands | Expected loss | One standard deviation | Approximate chance of being ahead |
|---|---|---|---|
| 100 | $12.50 | $285 | About 48 percent |
| 1,000 | $125 | $901 | About 45 percent |
| 10,000 | $1,250 | $2,850 | About 33 percent |
| 52,000 | $6,500 | $6,498 | About 16 percent |
The last row is the one to sit with. Solve for the point where expected loss finally equals a single standard deviation and you get about 52,000 hands. At 80 hands an hour that is roughly 650 hours at the table. Below that, the noise is larger than the signal, and any conclusion drawn from your results is a conclusion about the noise. The percentages in the last column are normal approximations and run slightly low at small hand counts because blackjack’s payout distribution is skewed by naturals and doubles, but the shape is right.
Two hands at once does not smooth anything out
Playing more than one hand at a time raises the volatility per hand rather than lowering it, because hands at the same table share a dealer card and are correlated. Published figures put the variance at 1.303 per hand on one spot and 1.782 per hand across two simultaneous spots, lifting the per-hand standard deviation from 1.142 to 1.335.
None of this means results cannot be questioned. It means a losing session, a losing week or a losing month at a 0.5 percent edge is exactly what the math predicts will happen constantly, and is evidence of nothing at all. The questions actually worth asking about game integrity, and the evidence that answers them, are set out on our page on whether online blackjack is rigged, along with what dealing and shuffle procedure should look like in shuffling and shoe penetration. On that last point, note that a continuously shuffled game and a cut-card game with identical posted rules do not have identical edges, and that constant reshuffling removes the composition changes that give card counting its foothold in the first place.
Checking the Rules Before You Sit
Online, the rules are required to be available to you. Check them in this order, because it is the order of how much money is at stake.
New Jersey’s standard at N.J.A.C. 13:69O-1.5(j) requires that enough information to identify the game, the game play and payout rules, and all charges imposed be readily available through the client before play begins and at all times during play, and specifically bars rules that depend on sound to convey their meaning. In the live game the equivalent obligations sit in the state’s table game regulations. Pennsylvania’s blackjack chapter, 58 Pa. Code Chapter 633a, runs from card values and shuffle procedure through insurance, surrender, doubling and splitting, and its payout section at 58 Pa. Code 633a.13 requires winning blackjack wagers to be paid at 1 to 1 with the exception of a blackjack, which is paid at 3 to 2, and insurance at 2 to 1.
The payout on a natural
3 to 2 or nothing. Worth 1.39 points and it takes two seconds to confirm.
Soft 17
Stands or hits, worth 0.22. Usually stated on the felt graphic or in the first line of the rules panel.
Double after split
Worth 0.14 and frequently omitted from marketing copy.
Doubling restrictions
Any two cards, 9 through 11, or 10 and 11. Worth up to 0.18.
Decks
Worth up to 0.48 in principle but rarely more than 0.06 in practice, since almost everything online is six or eight.
Surrender and resplitting aces
Worth 0.07 and 0.08. Nice to have, not deciding factors.
Peek or no hole card
Worth 0.11, and it changes your doubling decisions as well as the edge.
Side bet pay tables
Not part of the base edge at all, and capable of costing more than every rule above combined.
Which rules you can find on offer depends heavily on where you are, since the licensed catalog differs from state to state and the offshore market is a different question entirely. Our state-by-state guide to online blackjack covers where licensed play is available and what is on the tables there.
Last point, and the one the arithmetic on this page keeps circling back to: the house edge is the smallest of the three numbers that determine what blackjack costs you. Bet size and hand rate are the other two, and both are set entirely by you.
The rest of the fair play section
Seven guides sit alongside this one, each taking a single piece of the fairness question to its primary source.
Figures on this page were checked and the page was last reviewed on Aug. 25, 2026. Rule effects are drawn from published combinatorial analyses and are stated against the benchmark rule set defined above. Confidential help is available at any hour on 1-800-697-3738, keyed as 1-800-MY-RESET.
