How to Play Pai Gow Tiles and Set Both Hands

To play Pai Gow Tiles, identify your four tiles, check for ranked pairs and special combinations, then divide them into two hands of two tiles. Label the stronger one the high hand and the other the low hand. For ordinary hands, compare points, then the highest-ranked tile if points tie. Evaluate both hands separately: your high hand plays against the banker’s high hand, and your low hand against the banker’s low hand. No setting guarantees a win.

Pai Gow Tiles uses 32 Chinese domino tiles, with four dealt to each participant. High and low describe relative strength, not front or back positions on the table. Unlike seven-card Pai Gow Poker, this game depends on tile shapes, pair rankings and point-scoring rules.

A copy hand goes to the banker, who may be a player. Equal points therefore do not automatically produce a tied hand or a whole-wager push.

Which tiles and pairs should you learn?

Learn all 16 pairs in rank order: pip totals do not determine rank. Coordinates such as 1–3 describe the dots on the two halves of one tile; 1–3 + 1–3 describes a two-tile hand.

Eleven duplicated shapes account for 22 tiles. The remaining ten tiles complete the set: two form the supreme Gee Joon pair, and eight form four mixed pairs. Spellings vary between sources, so coordinates are the reliable way to identify tiles.

Official rules excerpt showing the Chong 3–3 tile and its individual rank

Original English illustration of Chong 3–3 from Hollywood’s June 2025 rules, page 1 of the Pai Gow Tiles section and page 346 of the complete PDF. “Rank 6” identifies the individual tile; the matched Chong pair ranks seventh in the pair table. Source: official PDF.

The rankings are supported by MGC’s rules dated 8 October 2020, §2 and Hollywood’s official Pai Gow Tiles section. These document specific rules, not universal conditions at every venue.

How do you determine a hand’s strength?

Check the combination’s category before calculating ordinary points. A pair, Wong or Gong outranks every ordinary point hand, regardless of its arithmetic total.

A Teen-based Wong outranks a Day-based Wong; the same applies to Gong. Remember that Yun 4–4 is also an eight and can make Gong.

Teen or Day with a seven makes a high nine, but it remains an ordinary nine-point hand—not a separate category above Gong. Always check pairs and special combinations before applying modulo ten.

How do ordinary points and separated Gee tiles work?

For an ordinary hand, add the two tiles’ pip totals and keep the final digit. This means discarding complete tens: 14 scores four, 18 scores eight, and 20 scores zero.

When separated, either Gee tile—1–2 or 2–4—may count as three or six, whichever produces the higher final score.

Together, the Gee tiles form the supreme pair because of the ranking rule, not because their arithmetic is flexible.

Their individual ranks are weak, but treatment varies. The common Wizard rules place both at the bottom; MGC’s historical rules distinguish them and their chosen values. Do not assume Gee is universally the lowest individual tile—check the table’s rules.

How do ties and copy hands work?

Compare category and pair rank first, then ordinary points, then the highest-ranked individual tile. If that higher tile also ties, the banker wins a copy hand; the second tile does not break the tie.

In the common order described by Wizard’s rules, individual tile ranks follow the pair list with Gee moved to the bottom. Mixed tiles with the same pip total share a rank. Individual Gee treatment can differ under other rules.

Equal points, but not a copy: Teen 6–6 with Chong 3–3 totals 12 + 6 = 18, scoring eight. Mooy 5–5 with mixed eight 3–5 totals 10 + 8 = 18, also scoring eight. The first hand wins because Teen ranks above Mooy.

A copy despite different second tiles: Bon 2–2 with Tit 1–6 totals 4 + 7 = 11, scoring one. Bon 2–2 with mixed seven 3–4 also totals 4 + 7 = 11. Both have Bon as their highest-ranked tile, so the banker wins the copy. The different second tiles do not matter.

For zero against zero, the bank wins without comparing highest tiles.

Losing one hand as a copy does not by itself settle the wager. The other hand still matters.

How do you check all three hand settings?

Enumerate all three ways to divide four tiles into pairs, then evaluate both hands in each split. This is a reproducible checking method, not an automatic optimal-strategy solver.

1. Label the tiles A, B, C and D, recording their coordinates.
2. Write the splits AB/CD, AC/BD and AD/BC.
3. For each hand, check category, ordinary points and highest-tile rank.
4. Label the stronger hand high and the other low.
5. Reject a split that is no better in either hand and worse in at least one.
6. Evaluate the remaining alternatives against a specific dealer policy and player strategy.

Maximizing the low hand is a useful starting heuristic, not an unconditional optimum. Improving it may weaken the high hand, and totals alone cannot establish the value of that trade-off.

Worked example: eight/four or seven/five?

For this set, one/one is weaker in both hands than either alternative. Choosing between eight/four and seven/five trades high-hand strength for low-hand strength.

Use these four tiles:

• A: Mooy 5–5 — 10 pips.
• B: Gor 1–3 — 4 pips.
• C: Foo 5–6 — 11 pips.
• D: Mixed seven 3–4 — 7 pips.

None of the resulting hands is a pair, Wong or Gong, so ordinary final-digit scoring applies. These are exact integer calculations; no rounding is needed.

In the one/one split, the Gor hand is the high hand because Gor outranks Mooy. That does not rescue the split: both hands lose to the corresponding hands in either other setting.

These examples were calculated independently, not copied from strategy tables. They show how to check scores and eliminate a dominated split. They do not establish a universal expected-value winner between eight/four and seven/five.

When should you split a pair?

“Never split a pair” is not a reliable rule: keeping a powerful pair may leave a very weak second hand. With two real pairs, putting one pair in each hand is a sensible starting point. With one valuable pair, check the remaining hand and applicable exceptions.

Consider Gee 1–2, Gee 2–4, Chong 3–3 and Gor 1–3.

Each separated Gee counts as three here, producing its better score. Counting it as six would produce two and zero respectively.

The published Wizard player strategy recommends splitting Gee Joon to make 7/9 or better. That is a player-strategy recommendation, not a universal house way or a complete optimal-play solution.

Keep three concepts separate:

• Legal setting: a split permitted by the applicable rules.
• Player strategy: advice on which legal split to choose.
• House way: the venue’s prescribed dealer-setting policy.

Hollywood’s June 2025 dealer policy uses different Gee-splitting conditions. MGC’s 2020 default dealer setting keeps the supreme pair. A dealer is not obliged to follow a published player strategy.

What should you check at the table, and how should you practice?

Confirm local scoring and settlement rules before playing, then practice all three splits on paper. Knowing the house way tells you how the dealer must set a hand; it does not reveal hidden tiles.

MGC’s rules dated 8 October 2020, §7 permit asking the dealer for setting assistance. This is a provision in dated rules; confirm whether and how assistance is available at your table.

For practice, copy the tiles from a worked example, hide the answer, list all three splits and compare both hands. Then apply a verified dealer policy. This checks your understanding, not the optimality of your strategy.

Side bets have separate economics and are outside the base-game house-edge figures below.

How does commission change the payout?

With a 5% commission on winning profit, a winning $25 stake produces $23.75 net profit and $48.75 returned including the original stake. This illustrates one commission structure, not every venue’s fee policy.

Here, 5% of $25 is exactly $1.25, so no additional rounding is needed. Check the actual commission basis and rounding convention before playing. Commission is not the entire house edge.

This is not a universal California rule. Player-dealer procedures and local fees may differ from casino-banked play.

Banking changes who benefits from copies, but may require enough funds to cover other wagers and may involve local fees and limits. It is not an unconditional recommendation or a promise of positive returns.

Which house-edge figures can you use?

The published Wizard figures of 2.44% and 1.66% support a limited comparison under specified assumptions. They are model values, not provider RTPs, measured session results or universal Pai Gow Tiles figures.

Wizard’s own analysis models non-banking play against Foxwoods house way with a 5% commission:

At $25 per round for 100 rounds, turnover is $2,500. Expected loss is $61 at 2.44% or $41.50 at 1.66%. Commission is already included: do not add another 5%. These calculations use the published model values directly.

The complementary 97.56% is the nominal expected return, or RTP, in the 2.44% input model. It is not the percentage of rounds you win.

Actual results vary. Expected loss is neither a safe bankroll nor a loss limit. See our house-edge guide for the concept and bankroll management guide for budgeting.

How do you calculate expected loss with ToolsGambling?

The house edge calculator calculates expected losses from manual inputs. It does not select tiles or evaluate your hand setting.

To reproduce the example:

1. Check “Enter a custom house edge” and enter 2.44.
2. Set “Bet Size per Round” to 25.
3. Select “By rounds”.
4. Set “Number of Rounds” to 100.
5. Only for hourly outputs, set “Rounds per Hour” to 30.
6. Only for long-term outputs, set “Sessions per Week” to 1.

Thirty rounds per hour is an assumed pace, not a measurement. It is not needed to calculate expected loss over the specified 100 rounds.

English calculator inputs showing a 2.44% house edge, a bet size of 25 and 100 rounds

Local capture of the English interface with inputs for 100 rounds at $25 per round and a 2.44% house edge.

Do not use the Pai Gow Poker preset for Tiles. The generic session simulator, slot RTP catalog and poker equity tool are not Pai Gow Tiles setting solvers either.

How should you interpret the calculator result?

The $61 expected loss is the modeled average cost of $2,500 in turnover at the entered house edge, not a prediction of your session loss. The $2,439 expected return is the expected total returned across that turnover, not a guaranteed ending bankroll.

English calculator results showing 100 rounds, 2500 wagered, 61 expected loss and 2439 expected return

Local capture of the English interface with the same parameters: 2.44% house edge, nominal RTP 97.56%, $2,500 wagered, 100 rounds, $61 expected loss and $2,439 expected return.

Changing the edge to 1.66 compares player strategies within the same published Foxwoods model, giving $41.50 expected loss. Entering that number does not prove that you play optimally.

The calculator does not simulate paired hands or pushes, estimate variance, calculate profit probability or assess risk of ruin. Use it to estimate the mathematical cost of a specified volume of play. Check tile settings separately: category, points, highest-ranked tile and both hands in each of the three splits.