Pair of 3s against a dealer 2
This is one of the closest decisions in blackjack. Split is correct, but it beats hit by 0.26% of your bet — and of the 350 hand-versus-upcard spots on the chart, only 4 are closer.
Worth being blunt about what that expected return means: at -13.8%, you lose money on this hand in the long run whatever you do. Playing it correctly does not make it a good spot — it makes it a 0.26% less bad one. The choice is between two losing options, which is exactly why it gets argued about.
Every option, ranked
| Action | Expected return | Cost vs. best |
|---|---|---|
| Splitcorrect | -13.82% | — |
| Hit | -14.08% | −0.26% |
| Stand | -29.28% | −15.46% |
| Double | -56.41% | −42.59% |
Expected return per unit staked, from an exact recursive engine rather than a copied chart. Figures are for Dealer Stands on Soft 17, DAS, pays 3:2. Actions not shown are not legal here.
It genuinely depends on the house rules
“It depends” is usually a dodge. Here it is literally true: under 1 of the 6 rule sets modelled, the correct play is something other than split.
| S17 · 3:2 | Split | -13.8% |
| H17 · 3:2 | Split | -13.5% |
| S17 · Surrender | Split | -13.8% |
| S17 · No DAS | Hit | -14.1% |
| S17 · No Peek | Split | -13.8% |
| S17 · 6:5 | Split | -13.8% |
Why it is close
Standing hands the hand to a dealer showing 2, who — having already checked and not got blackjack — busts 35.4% of the time. Two bad numbers close together is the whole story — the decision is a choice between them, and neither rescues the hand. The bust odds page has both curves in full.
What the count does to this
Split is correct at a neutral shoe, but not at every shoe. The two actions are worth exactly the same at a Hi-Lo true count of −0.41, so past that point hit is the better play — a deviation conventionally written as an index of 0.
Where that number comes from, and what one card of each rank is actually worth, is on the card counting page.