Where these numbers come from — and their limits
The statistics below are drawn from ilovemahj.com’s published “Mah Jongg Statistics” series (platform play, including human and bot games, across specific NMJL card years 2022–2024) and the strategy framework at Mahjong4Friends. They are real, published aggregates — not invented — but they are also platform data, not universal NMJL law:
- The card changes every year, so exact percentages shift from one card to the next.
- Online-platform results (with bots) may differ from your in-person table.
- Coaching observations in this article are compiled community experience, not official rulings.
- The current official NMJL card and your table’s house rules always win over any number here.
Use these figures to calibrate intuition — not as fixed odds to bet on. See Sources for the primary links.
What the data says: your baseline numbers
Before any “advanced” idea, know the benchmark. If you don’t know your real win rate, you can’t tell improvement from noise.
Your “fair” win rate is about 22.3%
On the ilovemahj platform, 89.3% of games end in a win and only 10.7% in a wall game. With four players of equal skill, your expected share of those wins is 25% — so your baseline expectation is 89.3% × 25% ≈ 22.3%.
The takeaway
If your long-run win rate sits above ~25%, you beat equal-level opponents. If it sits below 20%, your strategy has a systematic flaw — not just bad luck.
The East seat is worth about +10.82%
| Seat | Loss | Wall | Win |
|---|---|---|---|
| Non-East | 68.04% | 10.07% | 21.89% |
| East | 65.67% | 10.07% | 24.26% |
Source: ilovemahj.com “Statistics (part 2)”. East starts with one extra tile and moves first, which is worth a measurable edge. Since East rotates, the advantage evens out over a full round — but within a single game, play more aggressively in the East seat and more conservatively elsewhere.
The Charleston: information warfare
The Charleston is the only moment opponents hand you information directly — and most players give away more than they intend.
The “stop the Charleston” paradox
Stopping the Charleston is controversial. The platform data shows a striking split between bots and humans:
| Group | Win if they DON’T stop | Win if they DO stop | Change |
|---|---|---|---|
| AI / bots | 22.46% | 57.82% | +157% |
| Humans | 21.80% | 21.10% | −3.2% |
Source: ilovemahj.com “Statistics (part 2)” (online platform, includes bots). Bots stop only when hand completion crosses a math threshold; humans often stop on a false sense of security — “I have a few good tiles” — and then miss the补牌 (replenishment) the second Charleston would have given them.
You have a real read
- 8–9 of your tiles already point at one or two specific hands.
- Your remaining gaps are high-circulation tiles (middle numbers, common winds).
- Otherwise, keep passing — the expected value of more information is higher.
What they tell you
- What you receive = what that opponent does not need.
- What you passed that gets kept = what they are collecting.
- Wind-passing patterns reveal direction: a player dumping winds is unlikely to be on a Winds/Dragons hand.
A rule-legal trick worth knowing: on the First Left you may pass back to the same player a tile you received on First Right (a “return pass”). It cleans your hand and clouds their read. Full mechanics in The Charleston.
Hand-selection math: chase probability, not size
The biggest amateur error is over-pursuing a “tidy” or high-scoring hand. The data rewards the hand most likely to land — not the one worth the most points.
Edge digits beat middle digits
In consecutive-run hands, runs anchored on 1 or 9 succeed far more often than runs through 3 or 6. From ilovemahj’s card-year analyses (Consecutive Run, Line 5, one-suit version):
| Run | Success rate |
|---|---|
| 1-2-3-4 | 20.6% |
| 6-7-8-9 | 20.6% |
| 4-5-6-7 | 15.4% |
| 3-4-5-6 | 12.9% |
The same shape repeats across every run hand ilovemahj analyzed: low and high ends win most, the middle (especially 3 and 6) wins least. The reason is a resource-competition effect — middle digits serve more hand types, so everyone contests them; edge digits serve fewer types, so more copies survive in the live pool.
Apply it
- When choosing between a 1 and a 6 as your pair base, keep the 1.
- For consecutive runs, prefer 1-2-3-4 or 6-7-8-9; avoid 3-4-5-6.
- In 13579 hands, the low version (1-3-5) outperforms the high version (5-7-9).
Pattern difficulty is real — and measurable
ilovemahj’s 2024 pattern analysis shows which structures actually win. The flexible, joker-friendly ones dominate:
| Pattern (grouped) | Share of wins |
|---|---|
| 2 Kongs + 2 Pungs | 57.96% |
| 1 Kong + 2 Pungs + 2 Pairs | 17.36% |
| 3 Kongs + 2 Singles | 6.12% |
| … down to … | <1% each |
| Singles & Pairs family | least successful |
Takeaway, echoed by coaching experience: AI picks for maximum win rate, not maximum points. Singles & Pairs looks “neat” but depends on natural singles/pairs the joker cannot substitute — so it wins least. Only chase a hard, high-value hand when your tiles give you a clear structural edge (e.g. 4+ jokers).
Attack–defense switching
The hardest and most valuable skill is knowing when your hand is dead — and flipping from “I want to win” to “I will not deal in.”
Decode the exposures
- An exposed pung locks that suit/number into the player’s plan — and tells you what they are not doing.
- All one suit exposed → likely a same-suit or one-color hand. Two pungs of 2 and 4 → probably a 1-2-3-4 run in progress.
- After an exposure, tiles adjacent to it (e.g. 7/9 around an exposed 8) rise in danger.
Danger assessment by stage
| Stage (wall remaining) | Safer discards | Dangerous discards |
|---|---|---|
| Early (>60) | Tiles many have already discarded; obvious off-suit isolates | Suits no one has discarded; tiles adjacent to exposures |
| Mid (30–60) | Tiles 2–3 players have thrown; suits unrelated to exposures | Same suit as an exposure; numbers not yet seen |
| Late (<30) | Dead tiles (3 already shown); anything unrelated to any exposure | Any tile logically connected to an exposed hand |
Signs your hand is dead — switch to full defense
- 2–3 of your key tiles are already discarded or exposed (especially the pairs in a Singles & Pairs hand).
- Your direction collides with 2+ exposed players — too much competition for the same tiles.
- You hold many dangerous tiles and no safe discard.
- Wall <20 and you are still >4 effective tiles from a win.
In full defense, stop chasing. Throw “warm” (already-discarded) tiles, and prefer breaking your own pairs over feeding a likely winner. Full method in Defensive play.
Joker management
Jokers are the “nuclear option” of American mahjong — but more is not always better.
| Jokers held | Loss | Win | Wall |
|---|---|---|---|
| 0 | 90.2% | 4.6% | 5.2% |
| 1 | 78.2% | 13.2% | 8.6% |
| 2 | 61.0% | 27.0% | 12.0% |
| 3 | 43.8% | 41.1% | 15.0% |
| 4 | 29.9% | 51.4% | 18.7% |
| 5 | 20.1% | 55.7% | 24.2% |
| 6 | 13.4% | 52.0% | 34.5% |
| 7 | 10.3% | 46.6% | 43.1% |
| 8 | 0% | 66.7% | 33.3% |
Source: ilovemahj.com “Mah Jongg Statistics” (original). ilovemahj notes that 6+ jokers are rare, so the far-right figures are less statistically solid — and suspects a “paradox of choice” where too many jokers open too many options and the player loses the thread.
2 is the break-even
- 0–1 joker: pick low-joker-dependence hands; avoid Quint hands.
- 2–3 jokers: standard play; use them to complete pungs/kongs, keep flexibility.
- 3–4 jokers: the high-probability zone — push to win fast.
- 5+ jokers: win rate can fall as jokers crowd the hand; don’t over-reach for a big hand.
Jokers cannot be pairs
- A joker may stand in for a tile in a pung, kong, quint, or sextet — never for a pair or a lone single.
- So a Singles & Pairs hand barely benefits from jokers; a pung/kong/quint hand benefits most.
- Track exposed jokers: if you draw the natural tile, you may swap it for an opponent’s exposed joker — but exposing jokers early makes them a target.
The mental game
Beyond the tiles: reads, opponent types, and information hiding — the layer coaching experience keeps coming back to.
Body language is a hint, not proof
- A flash of delight after a Charleston pass → you just fed them something they wanted.
- Long hesitation after a draw → the tile relates to their hand; they are reconsidering direction.
- Instant exposure → that tile was core to their plan; they had waited a long time.
Adjust your plan to them
- Aggressive: exposes early, chases speed → defend, use their tells against them.
- Conservative: rarely exposes → you can build more freely; less pressure to fold.
- Stubborn: sticks to one hand past the point of death → control key tiles, force the dead end.
- Flexible: pivots constantly → hard to read; just protect yourself, don’t try to “read them dead.”
Hide your own information
Delay exposure even when you could call; keep a concealed hand guessing. In the Charleston, a “false move” — keeping a useless tile someone passed you — can mislead them about your direction. Early discards of edge tiles near your real target can bait a wrong read.
Training path: from data to intuition
Mastery is just these habits, repeated until they are automatic.
| Stage | Goal | How |
|---|---|---|
| 1 · Automate the card | Recognize any hand in <3s | Daily “card speed-read”: random tiles → name possible hands |
| 2 · Charleston decisions | Feel “hand completeness” | Log post-Charleston state vs final result; calibrate |
| 3 · Probability | Estimate live tiles in real time | Track the wall: count seen tiles, estimate remaining needs |
| 4 · Attack–defense | Catch “hand is dead” early | Replay each turning point: which discard dealt in? Could you have folded sooner? |
Post-game review checklist
- Did I lock a hand too early in the Charleston — and miss opponent signals?
- Did I pick the highest-probability hand, or the one that just felt best?
- Did I miss a joker swap — or expose jokers someone then took?
- Did I keep attacking after the hand was dead?
- Is my real win rate above ~22.3%? Is my East seat better than the rest?
These review habits are 麻将竞技通用原则 · General competitive principle (本站整理) — practical coaching experience, not official NMJL rulings.
Sources & further reading
Primary, verifiable data behind this article.
- ilovemahj.com — Mah Jongg Statistics (89.3%/10.7% wall split, 22.3% expected win rate, joker win-rate curve).
- ilovemahj.com — Mah Jongg Statistics (part 2) (East-seat +10.82%, consecutive-run success by digit, Charleston-stop analysis).
- ilovemahj.com — 2024 Card Statistics (pattern win-share: 2 kongs + 2 pungs = 57.96%).
- Mahjong4Friends — American Mah Jongg Strategy (hand-selection and exposure framework).
- Coaching-community observations on tells, opponent types, and information hiding are compiled experience, not attributed to any single player.
All percentages are platform aggregates for specific NMJL card years and may not match the current card or your local table. The official NMJL card remains the final authority on every hand. Full site bibliography in Sources.
Where to go next
This analysis is the “why.” These pages are the “how.”
For the rules underneath all of this, see Rulings & case library, Joker rules, and the Sources page.