actfap

Pai Gow Poker Hand-Setting Trainer

Deal seven cards, set your hand, then see the house way and the cost of your choice against best play.

This prices every legal way to set a dealt seven-card Pai Gow Poker hand against one published house way, by simulation. It does not model a specific casino's exact commission structure, a commission-free variant, or any side bet.

Worked example

Two pair, set with both pairs kept together in the back hand. Deal a hand to set one yourself.

Dealer’s hand (house way)

Front
Ace High
Back
Pair of Eights

Example hand

Front
Queen HighLoses
Back
Two Pair, Kings and SevensWins

What this setting cost

Pricing every legal way to set this hand.

The one hand dealt

Push

You beat the dealer on one hand and lost (or tied) the other. No money changes hands on a push.

Every legal way to set this hand, ranked by simulated value

Pricing every legal way to set this hand.

Pai Gow Poker pushes on a large share of hands by design; a low per-hand cost does not mean the game pays back more than it takes. The 5% commission modeled here is charged on wins, which is the standard convention, not every table’s exact rule. Nothing on this page identifies a hand worth chasing: every legal setting shown above has a negative expected value against the house way.

Written and reviewed by Alexander Korsager · last reviewed October 3, 2026 · corrections

The idea

What this tool shows

Pai Gow Poker's whole decision is how to split seven cards into a five-card hand and a two-card hand, with one rule that never bends: the two-card hand can never out-rank the five-card one. Every tool that covers this game answers that question the same way, enter seven cards, get the one setting a computer thinks is best. This one deals the hand to you, lets you set it yourself, and then shows the cost of the choice you actually made, not just whether it matched the answer key.

  1. The dealer never chooses.A fixed, published set of rules called a house way decides for them, the same house way used at Foxwoods and reproduced here in full. A player is free to set their own hand any legal way they like, and the decision is a real one: the same seven cards can be set several different legal ways, and they are rarely worth the same amount.

  2. Every setting shown on this page is priced by simulation, thousands of dealer hands drawn and set by the same house way, with every candidate setting judged against the identical sequence of dealer hands rather than separate ones. That is what makes it possible to say a setting cost you a specific amount rather than just that it was wrong: the gap between two settings is measured far more precisely than either setting's own value is, because most of the noise in what the dealer draws cancels out of the comparison. Where that noise still cannot rule out a tie, this page says so instead of printing a number that might be backwards.

  3. Nothing here identifies a winning hand.Every legal way to set every hand carries a negative expected value against the house, the same way every hand in blackjack does before you count cards. What this measures is how much worse some settings are than others, not whether any of them make money.

FAQ

Questions players ask

What is a house way in Pai Gow Poker?

A fixed procedure the dealer follows to set their own seven-card hand, so every dealer at a given casino sets identical cards identically. It exists because a live dealer cannot make a judgment call on every hand without slowing the table, and because a fixed rule cannot be exploited the way a dealer's own preference could be. House ways differ by casino and by dealing-software vendor. This page uses Foxwoods' published house way, one of several documented versions.

Can the two-card hand ever beat the five-card hand?

No. That is the one rule that is never optional: if your two-card hand would out-rank your five-card hand, the setting is not legal and this page will not let you lock it in. A pair of nines in front and a busted five-card hand in back, for instance, is illegal, because a pair out-ranks no pair.

What does the joker do in Pai Gow Poker?

It is often called the bug because it is not fully wild. It can stand in as an ace, or as whichever card completes a straight, a flush or a straight flush, and nothing else. It cannot turn three kings into four, and it cannot make an arbitrary pair. Four real aces plus the joker is five aces, the single highest hand in the game, ranked above a straight flush.

Why does the dealer win a tie?

In Pai Gow Poker, an exact tie between the player's hand and the dealer's hand on either the front or the back, called a copy, always goes to the dealer. That rule alone is a meaningful part of the house's edge, on top of the commission charged on a win.

How is the EV cost of a setting calculated?

By simulation, not a formula: thousands of dealer hands are drawn and set by the house way, and every legal way you could have set your own seven cards is judged against that identical sequence of dealer hands. Comparing two settings against the same draws, rather than separate ones, is what lets this page state a gap with real confidence at a trial count that resolves in a fraction of a second. See this tool's methodology for the measurement behind that trial count.

Is there one correct way to set every hand?

Usually, but not by a wide margin on most hands. A hand with an obvious pair and nothing else has one clearly better setting. A hand with two ways to make a decent front tends to be much closer, and this page's withholding rule exists for exactly that case: where the simulation cannot tell two settings apart with confidence, it says so rather than asserting an order that sampling noise could have gotten backwards.

Methodology

How this is calculated

Every legal split of seven cards into a two-card front hand and a five-card back hand is priced by drawing thousands of seven-card dealer hands from the remaining pack, setting each one by the house way, and recording how the player's setting fares against that same sequence of dealer hands (win, push, or lose, with a 5% commission charged on a win).

Because every candidate setting is judged against the identical sequence of dealer draws rather than separate ones, the standard error on the difference between two settings is far tighter than either setting's own standard error, which is what lets this page state a confident gap at a trial count that runs in a fraction of a second.

Common misreadings3
  • A setting with a less negative expected value is still a losing setting. Every legal way to set every hand carries a negative expected value against the house; nothing on this page identifies a hand worth chasing.
  • "Statistically tied" does not mean the two settings are equally good in an absolute sense, only that this simulation cannot tell which is better with confidence. One may still be better in reality.
  • The single dealt outcome shown after you set a hand (win, push or lose) is one draw from a random deck, not a measure of how good your setting was. Read the ranked list and the EV gap for that, not the one outcome you happened to draw.
Assumptions8
  • The dealer's hand is set by Foxwoods' published house way (Wizard of Odds, retrieved 2026-09-14), reproduced in full in this tool's source. Other casinos and dealing-software vendors publish different house ways; a real table's dealer may set some hands differently from this page.
  • A 5% commission is charged on a win, the conventional rate. Some tables run a commission-free variant with different payout rules on specific hands; that variant is not modeled here.
  • A tie on either hand (a copy) always goes to the dealer, per standard Pai Gow Poker rules.
  • 3,000 dealer hands are drawn per simulation, using paired sampling: every candidate setting is judged against the same sequence of draws. A companion measurement (this tool's own build notes) found the gap between the best and second-best setting typically clears several standard errors at this trial count when sampled this way, against roughly an order of magnitude more trials an independently-sampled comparison would need for the same confidence.
  • One hand-setting situation in the source house way is genuinely ambiguous or under-specified in the published text: the written exception for an ace held alongside a pair of kings, queens, jacks or tens when a straight, flush or straight flush is also achievable, which trades the made hand away entirely but does not say which card joins the pair or what the back hand becomes. This page keeps the hand category the house way's own logic already committed to and maximizes the front hand within that constraint, rather than guessing at a case the source text does not spell out. A documented, deterministic choice rather than the one certain reading of the rule.
  • A hand needs all four real aces plus the joker, with no other card in the five-card grouping, to count as Five Aces (58 Pa. Code 645a.6: "a high hand consisting of four aces and a joker"). Three real aces plus the joker plus any fifth card is Four of a Kind, Aces, ranked well below a straight flush, not above one. A defect found and fixed 2026-09-16 had this page's evaluator granting Five Aces to that weaker hand too, whenever the joker completed three real aces into a foursome regardless of what the fifth card was; every Five Aces figure and every hand this page displays as Five Aces was regenerated after the fix, and math.test.ts now pins the corrected boundary directly against the regulation's wording.
  • A second defect, found and fixed 2026-10-02, had the dealer's Straight/Flush/Straight Flush branch always applying the generic "keep the made hand in back, maximize the front" rule, even on the seven-card hands that also carry two pairs overlapping the achieved hand's ranks. The published house way's own text names that composition separately ("With Two Pair: Play according to two-pair strategy") rather than folding it into the No Pair/One Pair wording the generic rule implements, and this page's code did not check for it. Fuzzed across 1,000,000 random seven-card hands, hands whose best achievable five-card hand was a straight, flush or straight flush made up about 11.4% of the sample, and about 5.3% of those also carried the overlapping two pair, around 0.6% of all hands dealt, never zero. The fix shares the existing Two Pair branch's logic (already implemented and tested on its own) with the Straight/Flush/Straight Flush branch rather than writing it twice, and math.test.ts now pins both the two-pair-overlap case and the superficially similar three-of-a-kind coincidence that the generic rule already handled correctly by construction.
  • A third defect, found live on this page and fixed 2026-10-03 (#1244), had every joker-completed pair, triple or four-of-a-kind side pair (outside the Five Aces and the joker-completed quad-of-aces case in the entry above) falling to a generic fallback instead of the house way's own named rule for the category the hand had actually reached. A joker pairing a real ace, for instance, was not recognized as a pair at all: a hand of joker, ace, queen, queen, seven, three, two (two pair, aces and queens) was being set with the strongest possible back hand and the weakest front, instead of the house way's own "big pair is aces: place the small pair in front" rule, queens to front and the pair of aces kept back. Fixed by counting the joker as an ace when grouping ranks for every category except where it is completing a straight, flush or straight flush, its only other legal role. Two further defects fixed in the same pass: a four of a kind in the 7s-to-10s range with the joker as the only card of at least King rank among the side cards was being split instead of kept whole, because the check compared the joker's literal rank field (which is 0, not an ace's 14) rather than its ace-high sort position; and the Straight/Flush/Straight Flush branch's "with no pair" rule was filtering candidate splits down to only the single highest-ranked straight or flush achievable, when the published rule allows any complete one in back, which could force a weaker front than the hand allowed. Re-simulated 300,000 player-versus-house-way hands against this page's own dealerSet on both sides at the conventional 5% commission: push rate moved from about 45.5% to about 41.6%, and the house's edge from about 2.42% to about 2.56%, both now closer to published Pai Gow Poker figures than the pre-fix numbers were. math.test.ts pins all three corrected cases and the deliberately unchanged quad-of-aces case next to them.
Limits3
  • This prices one dealt hand and one house way. It does not track a session, a bankroll, or how the cost of a setting compounds over many hands, which is the Bankroll Calculator and Variance Simulator's job, not this page's.
  • The withheld-gap rule is conservative by construction: it can say two settings are statistically tied when a much larger trial count would in fact separate them. That is the intended trade, not a defect: this page would rather understate its own confidence than assert an order that might be backwards.
  • A different casino's house way, commission rate, or commission-free ruleset will change both the dealer's setting and the exact cost of any player setting. This page prices one documented version, not every table you might sit at.
Evidence

The house way is transcribed verbatim from Wizard of Odds' Foxwoods Pai Gow Poker house-way page (wizardofodds.com/games/pai-gow-poker/house-way/foxwoods/, last updated 2026-08-03, written by Michael Shackleford, retrieved 2026-09-14), and every named branch (no pair, one pair, two pair by big-pair rank, three pair, three of a kind, two sets, straights/flushes/straight flushes including the two-pair sub-case, full house, four of a kind, four of a kind with a spare pair, and five aces) is implemented and unit-tested against a worked example.

The five-card and two-card evaluators, including the joker's ace/straight/flush-only role, are tested against standard poker hand rankings. The Five Aces boundary specifically is checked against 58 Pa.

Code 645a.6 ("Pai Gow Poker rankings"), retrieved 2026-09-16, which is the source that surfaced the 2026-09-16 defect described above.

Legality (the back hand is never out-ranked by the front) is fuzz-tested across 4,000 randomly dealt seven-card hands in math.test.ts, alongside every named house-way branch and the simulation's determinism.

The straight/flush/straight-flush-with-two-pair prevalence figure above is a separate, larger fuzz (1,000,000 hands) run against this page's own evaluator, not pinned in the unit suite since it is a measurement of frequency rather than of correctness.

Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.

Related tools

Try next

We’d like to use privacy-friendly analytics (Google Analytics, Vercel Analytics and Speed Insights) to see which tools and pages are actually useful, so we can improve them. Nothing is sold, and nothing loads unless you say yes. See our privacy policy for details.