August 5, 2026 · 7 min read · blackjack · splitting · basic strategy

When to split in blackjack: 2 always, 2 never, 6 maybe

Always split aces and 8s, never split 10s or 5s, and split 2s, 3s, 4s, 6s, 7s and 9s only against the dealer upcards where the math says yes. In the standard American game where the dealer peeks for blackjack, that is the entire answer, and no reputable source disagrees with any part of it (no-hole-card games are the one exception, and we will get to it). What almost nobody explains is why — which is the part that makes the rules stick when a dealer is waiting and the table is watching.

Splits come up more often than they feel like they do. Add up the pair rows in the Wizard of Odds' six-deck expected-return appendix and they total 0.138 — that column counts hands where you hold a pair and the dealer does not have blackjack, so the raw pair frequency is a shade higher at 0.145. Either way, about one hand in seven arrives as a splittable pair. Splitting is also the only decision in blackjack where the correct play sometimes means putting a second bet on a hand you fully expect to lose, and the only one where a genuinely good hand should be left alone. Both of those are why it gets misplayed.

Always: aces

A pair of aces is a 2 or a soft 12. Neither total wins anything on its own, and neither can be improved without drawing. Split them and each ace becomes the single strongest card in the deck: sixteen cards in every fifty-two are worth ten, so each half has close to a one-in-three shot at 21 on the very next card.

The Wizard of Odds publishes expected returns for every starting hand under six decks with the dealer standing on soft 17. Against a dealer 6, hitting a pair of aces returns +0.188 units per unit bet; splitting returns +0.682. Against a dealer ace — the worst upcard on the table — hitting returns −0.022, a small loss, while splitting returns +0.127, an actual profit. There is no upcard anywhere on the board where splitting aces is not the best available play.

Which is precisely why casinos fence the move in rather than the reverse. In most games each split ace gets exactly one card and nothing more: no hit, no double, often no resplit. An ace plus a ten after a split is a 21, not a natural, so it pays even money instead of 3-to-2. Those restrictions are not decoration — against the Wizard's eight-deck baseline, being allowed to draw to split aces is worth 0.19 percentage points to you, resplitting aces 0.08, and having ace-ten after a split count as a 3-to-2 blackjack 0.21. Being forbidden to split aces at all costs 0.18.

Always: 8s, including the one that hurts

A pair of 8s is a hard 16. Wikipedia's article on the play, citing Marten Jensen, calls 16 "the worst hand one can have in blackjack" — too low to stand against anything strong, too high to hit without a serious risk of busting. Splitting does not turn it into a good hand. It turns it into two hands of 8, each of which becomes 18 on a ten, and 18 is a total you can actually win with.

Against weak upcards this is not damage control, it is a profit. In the Wizard's six-deck table with double-after-split allowed, splitting 8s has a positive expected return against every dealer upcard from 2 through 7: +0.082 against a 2, +0.319 against a 7, +0.406 against a 6. Hitting that same 16 against a 7 returns −0.408. One decision, a swing of roughly three quarters of a unit. (Henry Tamburin's column describes the winners as running from 3 to 7 rather than 2 to 7; the Wizard's table has the 2 just barely in the black.)

Then there is 8,8 against a 10, the play people refuse to make. The numbers, same table: standing returns −0.537, hitting −0.535, splitting −0.475. Surrender, where offered, returns exactly −0.500. Splitting is the best of a bad set, and it stays the best even without double-after-split, where it returns −0.486. Tamburin put the same idea in dollars across the 9, 10 and ace upcards: hitting a pair of 8s loses about $52 per $100 wagered, splitting about $43. The Las Vegas Advisor's framing is the one to remember — against a dealer ten every option loses money, and splitting simply loses the least.

The honest footnote is that this is a small error. The Wizard has computed the cost of habitually hitting 8s against a 9, 10 or ace instead of splitting: 0.019 percentage points on the house edge. It matters less than which table you sat at. But it is free to get right, and getting it right is what stops you from also getting the profitable splits wrong.

Never: 10s, and why the dealer's 6 is the trap

Two 10s make 20, which wins on its own about as reliably as anything in the game. Splitting it is greed with a strategy chart's vocabulary, and the tell is that people only ever propose it against a bust card.

So take the friendliest possible case: a dealer 6. Standing on two 10s there returns 0.702826 units. The Wizard's Ask-the-Wizard answer puts splitting at 0.622165 and notes that standing therefore wins an extra 8.07 percentage points. His Appendix 9 table for the same game — the one that lets you resplit to four hands — puts splitting lower still, at 0.444942, because every extra hand you resplit 10s into compounds the same mistake. The two figures come from different Wizard computations of the same hand; the likeliest explanation is a difference in how freely each allows resplitting. Both lose to standing, and standing wins by more the more freely you are allowed to keep splitting.

Against a dealer 10 the gap stops being a debate. Standing on 20 returns +0.559; splitting returns −0.420. That is a hand worth more than half a bet traded for one that loses nearly half a bet, at a table where the dealer's own 10 finishes on 20 about 37% of the time. If you ever want a single sentence for the player at third base: 20 already loses to nothing but 21, and no upcard is weak enough to make two unknown hands worth more than that.

Never: 5s, because it is not really a pair

A pair of 5s is a hard 10, and a hard 10 is a doubling hand. The Wizard's chart does not even give 5s a splitting row — the instruction is to look the hand up as a hard total, the same as 10s.

The arithmetic is not close. Against a dealer 6: doubling returns 0.599, hitting 0.299, splitting 0.062 even with double-after-split allowed. You would be giving up the second-strongest doubling total in the game — only hard 11 is better — to start two fresh hands from a 5, which is the card stiff totals are built out of. The old table saying is that you never split anything beginning with F — fours, fives and faces. The Wizard points out it is true of two of those three, which is a useful reminder that catchy rules are how the one genuinely profitable split gets skipped.

The six conditional pairs

These are the lines to actually learn, straight from the Wizard's four-to-eight-deck basic strategy. DAS below means double after split is allowed.

Five of the six run on one idea: split into weakness, fold into strength. In a six-deck game where the dealer stands on soft 17, the dealer busts 42.3% of the time showing a 6, 41.8% showing a 5 and 39.6% showing a 4 — but only 22.9% showing a 9, 23.0% showing a 10 and 16.7% showing an ace. A second bet against a card that busts two times in five is cheap. The same bet against an ace is buying a second ticket to the same loss.

9s are the exception, and the exception is instructive. 9,9 is 18, a hand that is already ahead of most dealer outcomes, so splitting has to beat a decent baseline rather than rescue a bad one. Against a dealer 7 it fails: standing returns +0.400, splitting +0.364. The reason is mechanical — a dealer 7 finishes on exactly 17 about 37% of the time, and your 18 beats 17. Move the upcard to an 8 and 18 is no longer good enough (standing +0.099, splitting +0.230), so you split. Against a 10 or ace you stand again, for the blunter reason that two hands starting on 9 against a strong dealer is just two ways to lose.

7,7 against a 7 is the other line that looks wrong. Hitting a 14 there returns −0.332; splitting returns −0.049 with DAS. Neither makes money. But a 14 has to draw and often busts, while two hands starting on 7 get to build — and the dealer's 7 is the upcard least likely to reach 18 or better.

2s and 3s: split against a dealer 4 through 7. Add a dealer 2 or 3 to that list if DAS is allowed.
4s: split only when DAS is allowed and the dealer shows a 5 or 6. Otherwise hit.
6s: split against a dealer 3 through 6, plus a dealer 2 if DAS is allowed.
7s: split against a dealer 2 through 7, DAS or not.
9s: split against a dealer 2 through 6 and against an 8 or 9. Stand against 7, 10 and ace.
And the anchors: aces and 8s always, 10s and 5s never.

Double after split moves four of these lines

DAS is not a footnote. The Wizard's appendix gives the whole six-deck, stand-on-soft-17 game an expected return of −0.004026 with double-after-split and −0.005445 without — a gap of 0.14 percentage points, and his rule-variations table prices the loss of DAS at the same 0.14. Most of that value shows up in exactly four hands: 2s, 3s, 4s and 6s.

4,4 against a dealer 6 is the cleanest demonstration. Hitting returns +0.124. Splitting without DAS returns +0.018, which is worse than hitting. Splitting with DAS returns +0.168, which is better. Nothing about the cards changed; the permission to double the split hands did, and it flips the decision — because a 4 paired with a 6 or a 7 is precisely the total you want to put more money on. The same flip happens at 4,4 versus a 5, at 2,2 and 3,3 against a dealer 2 or 3, and at 6,6 against a dealer 2. Note 6,6 versus a 3 is not on that list: it is a split with or without DAS.

The Wizard names splitting 4s against a 5 or 6 with DAS as the play most likely to be misplayed at a real table, and says he has essentially never seen another player make it. Hitting instead costs 2.83 percentage points of expected value against a 5 and 4.38 against a 6. It is also rare — about one hand in 1,135 in a six-deck game — so playing it wrong every single time raises the overall house edge by 0.0032%. Both things are true: it is the split people get wrong, and it is worth almost nothing. Learn it anyway, because it is the line that proves you understand what DAS is for.

The rules that decide how much splitting is worth

Every figure above assumes an American game where the dealer peeks for blackjack before you act. That is the hidden reason splitting 8s against a 10 works at all: by the time it is your turn, you already know the dealer does not have 21, so the appendix values are conditional on that. In a European no-hole-card game the dealer takes the second card after you have played, and a dealer blackjack can collect both halves of your split. The Wizard prices that rule's effect on splitting at 0.03 percentage points.

Here the sources genuinely disagree. Wikipedia's article says that in a no-hole-card game you should not split aces against an ace, nor 8s against a 10 or ace — but both claims carry citation-needed tags. The Wizard ran the aces case for a one-deck no-peek game and came out the other way: hitting returns −0.533, splitting −0.223, on the grounds that being able to draw to the split aces outweighs the extra exposure. If you play no-hole-card games regularly, check your house's exact wording rather than trusting either statement.

The rest of the split economics are printed on the felt or available by asking the dealer. Against the Wizard's eight-deck, stand-on-soft-17, DAS, split-to-four baseline:

One mild disagreement worth flagging: Tamburin puts resplitting aces at about 0.06 percentage points rather than the Wizard's 0.08. The number is small enough that the practical advice is identical — when a casino offers it, take it.

Not being allowed to split at all: 0.57 percentage points against you. Not being allowed to double or split: 1.91.
No splitting aces: 0.18 points. No double after split: 0.14 points.
No resplitting: 0.10 points. Splitting to only three hands instead of four: 0.01 points.
Resplitting aces allowed: 0.08 points in your favour.
Drawing to split aces allowed: 0.19 points. Doubling or standing after splitting aces: 0.08 points.

Ten ranks, four of them unconditional

That is the whole subject. Ten pair ranks; four of them are answered before you look at the dealer; the remaining six are one line apiece, and four of those six move when the DAS rule moves. It is a smaller body of knowledge than it feels like from the far side of a strategy chart.

Knowing it and executing it are different skills, though, and the gap between them is speed. The decision trainer here deals hands and asks for hit, stand, double, split or surrender, then tells you why the correct answer is correct instead of just marking you wrong. There is a Pairs filter that serves nothing but split decisions, so you can drill all ten ranks in a few minutes rather than waiting for pairs to appear one hand in seven. And it reads your rule settings: switch double-after-split off and the correct answers for 2s, 3s, 4s and 6s change with it, which is the fastest way to feel what that rule is actually doing to your game.

Four splits need no thought and six need one line each. Drill them against a live dealer upcard until the pair does the deciding for you.

Drill split decisions free →

Share this post — the link unfurls with its own card.