Backgammon — Official Rules and Modern Guide
Backgammon is a 5,000-year-old race game with surprisingly deep tactics. Two players, 15 checkers each, two dice, and a doubling cube. This guide covers the official rules from the start of the game to the bear-off, plus the cube, scoring, and how modern AI engines fit into both casual play and serious analysis.
This is the guide to backgammon. Long narde has its own rules and its own article: long narde — official rules.
What is the object of the game?
Move all 15 of your checkers around the board, into your home board, and then bear them off. The first player to bear off all 15 wins. Depending on whether the loser has borne off any checkers and where their remaining ones are, the win is worth 1, 2, or 3 points — multiplied by the doubling cube if it's been turned.
Backgammon starting setup: how many checkers and where do they go?
In backgammon each player has 15 checkers, 30 on the board. The board has 24 points (long triangles) numbered 1–24, with each player counting from their own perspective. The starting setup splits those checkers across four points: 2 on the 24-point, 5 on the 13-point, 3 on the 8-point, and 5 on the 6-point — the 2-5-3-5 arrangement. The opponent's setup is mirror-symmetric.

How do the dice and movement work?
Each turn you roll two dice. The two numbers are independent moves — you may move one checker the sum if the intermediate point is legal, or two separate checkers each by one die. Doubles (e.g., 5-5) play four times. White moves toward the 1-point; Black moves toward the 24-point. The two players' paths cross in the outer board.
A checker may land on:
- An open point (no opposing checkers).
- A point with one opposing checker — this is a hit: that checker goes to the bar.
- A point with one or more of your own checkers — stacking is unlimited.
It may not land on a point held by two or more opposing checkers — that point is "made" by the opponent and is closed to you. If your roll has no legal play, you forfeit the turn. If only part of the roll is playable, you must play the largest legal portion.
How do hits, the bar, and re-entry work?
Sending a single opposing checker to the bar is the engine of backgammon's tactical layer. While a checker is on the bar, the player cannot make any other move until they bring it back into the opponent's home board. To re-enter, you roll dice; the number on each die must correspond to an open point in the opponent's home (1–6 for the hit player coming from White's bar, 19–24 from Black's bar). If neither die offers an entry, the turn is forfeited.

Closing all six home-board points is called a closeout: the hit player cannot re-enter at all. Maintaining a closeout while you bring builders home is one of the strongest positional weapons in backgammon.
How do you bring your checkers home and bear off?
Once all 15 of your checkers are in your home board, bear-off begins. Each die removes a checker from the corresponding point. If the rolled number is higher than your highest occupied point, you bear off the highest checker. If you have a checker on a higher point, you must move within the home board (you cannot waste pips that could be played on a higher checker).

If you get hit during bear-off, the hit checker goes to the bar and must re-enter, travel all the way back home, and only then can you resume bearing off. This is why you avoid leaving a blot on a point your opponent can roll to.
What is the pip count and how does it decide the race?
The pip count is the sum of pips each of your 15 checkers needs to travel before being borne off. Both players start at 167. The lower count is ahead in the race — but pip count alone doesn't decide the game while contact is possible. Once contact is broken (no further hits possible), the race becomes a pure dice-and-pip problem with well-understood probability tables. Modern engines compute pip and exact race equity instantly; humans use approximation rules like the cube-decision benchmarks.

How are points scored: single, gammon, backgammon?
- Single (normal) win = 1 point: you bear off all 15, your opponent has borne off at least one.
- Gammon = 2 points: you bear off all 15, your opponent has borne off zero.
- Backgammon = 3 points: gammon AND your opponent has at least one checker on the bar or in your home board.
These multipliers compound with the doubling cube: a gammon at cube value 4 is worth 8 points, a backgammon at 4 is 12, and so on. This is why pre-cube positional decisions often turn on whether you're playing for a single or trying to escalate to a gammon.
How does the doubling cube work in match play?
In money play and matches longer than one game, either side may turn the cube before their roll. The cube starts neutral at 1; the side accepting a double becomes the cube owner and may redouble in the future. The opponent who refuses concedes the current stake (before the double).

Crawford rule, post-Crawford recubes, the 64-cap, and Holland are all standard cube conventions. For the math behind take/drop decisions and the leader/trailer thresholds our engine uses, see the dedicated doubling cube article.
How do modern engines analyze backgammon?
Backgammon is one of the most heavily studied games for stochastic AI. TD-Gammon (1992) was a watershed: a neural network trained almost entirely from self-play that reached near-expert level. Today, gnubg and eXtreme Gammon are the two reference engines, with newer projects (Nardex, wildbg) bringing modern stacks (Rust, ONNX, GPU inference) and HTTP APIs.
If you build software that needs backgammon analysis — apps, coaching tools, content generation — see the Nardex vs gnubg vs XG comparison and the 5-minute API quickstart. Both are written for developers and cut to the technical detail.
Where to go next
- How the Doubling Cube Works — math of take/drop, Crawford, leader/trailer thresholds.
- Long Narde rules — the Russian variant, no hits, single head, mars/koks.
- Long narde vs short backgammon — 7 differences — cross-cluster comparison.
- Developers — engine API, beta access, deep technical pages.
- Play backgammon against the computer online — a game against the Nardex neural engine right in the browser.
Frequently asked questions
- What's the difference between backgammon and long narde?
- Backgammon has four starting stacks, bidirectional movement converging on each home board, hits and the bar, and gammon/backgammon scoring. Long narde uses a single starting head, unidirectional movement, no hits, and mars/koks scoring. See our cross-cluster comparison for details.
- How many checkers are there in backgammon?
- In backgammon each player has 15 checkers — 30 on the board. They start from four points: 2 on the 24-point, 5 on the 13-point, 3 on the 8-point, and 5 on the 6-point (the 2-5-3-5 setup). The opponent starts mirror-symmetrically.
- Do I have to use both dice on every roll?
- Yes — if any sequence of plays uses both dice, you must play it. If only one die can be played, you play that die and lose the other. If you can play either die alone but not both, you must play the higher one.
- When can I start bearing off?
- Only after all 15 of your checkers are in your home board (points 1–6 for White, 19–24 for Black). If a checker is hit during bear-off and sent to the bar, you must re-enter and bring it home again before resuming.
- What's a gammon vs a backgammon?
- A gammon is when you bear off all 15 checkers while your opponent has borne off none — worth double. A backgammon is the same condition AND your opponent still has a checker on the bar or in your home board — worth triple. Both are multiplied by the cube value.
- How does the doubling cube work?
- The cube starts at 1 and doubles to 2, 4, 8, 16, 32, 64. Either player may offer to double before their roll if they own the cube (or before the first double if neutral). The opponent accepts (taking ownership) or refuses (conceding at the current stake). See our doubling cube article for the math.
- What is the pip count?
- The pip count is the total number of points all your checkers must travel to bear off. Each player starts with a pip count of 167. The pip count is the cleanest metric for evaluating racing positions — lower is better. Engines compute it instantly; humans count it during pauses.
- Can I analyze my games with AI?
- Yes. Modern engines (gnubg, eXtreme Gammon, Nardex) evaluate positions and rank moves by equity. Nardex exposes this through an HTTP API; see our developers' quickstart and the comparison page for how we differ from gnubg and XG.