Klondike Solitaire, measured: how often can you really win?

This is Klondike, the solitaire most of us simply call "patience": seven columns, a stock pile, and four foundations to fill from Ace to King. The version on this page turns one card at a time from the stock (Turn 1), and you can go through the stock as often as you like.

So how winnable is it? We dealt 10,000 games to a computer player that sees exactly what you see, and it won 42.3% of them without ever pressing Undo. A second program that was allowed to peek at every hidden card won at least 76.6% of 1,000 deals, which tells you most of the losses are not the cards' fault. Somewhere between those two figures is where a careful human player lives.

Below you will find the numbers, the rules as they work on this site, and the handful of habits our data says actually move your win rate. Deal a game above whenever you are ready.

At a glance: Klondike Turn 1 in numbers

Our perfect-knowledge solver left 21.3% of deals undecided when it ran out of search budget, so our honest range for "winnable with perfect knowledge" is 76.6% to 97.9%. The published 90.48% sits comfortably inside it.

The rules of Klondike on this site

The deal

• Seven tableau columns holding one to seven cards. Only the bottom card of each column is face up.
• The stock: the remaining 24 cards, face down, top left.
• Four foundations, top right, empty at the start.

Goal

Move all 52 cards to the foundations, one pile per suit, from Ace up to King.

Moves you can make

• Build down the tableau in alternating colours: a black 6 on a red 7, a red Jack on a black Queen.
• Move any face-up run as a group, as long as its top card fits where it lands.
• When you move the last face-up card off a hidden one, the hidden card turns over.
• Only a King (or a run headed by a King) can go into an empty column.
• Click the stock to turn one card onto the waste pile. The top waste card can go to the tableau or a foundation.
• When the stock runs out, turn the waste back over and go again, as many times as you like.
• You may take a card back down from a foundation onto the tableau if you need it to catch another card.

Prefer turning three at a time? That is a very different game, statistically. See Klondike Turn 3.

What the numbers say about playing well

1. Most losses are decisions, not bad luck

The single most useful number on this page is the gap between 42.3% and 76.6%. Our no-peeking player follows a fixed priority list and never takes a move back. The perfect-knowledge solver shows that at least three in four of the same deals can be won. Only 2.1% of deals were proven impossible. When a game goes wrong, the odds say there was usually a better line. That is exactly what unlimited Undo is for.

2. Turn over hidden cards first, starting with the tallest column

After "play anything to the foundation", our computer player's top priority is any move that turns over a face-down card, starting with the column that hides the most. The reason is simple: 88% of deals begin with at least one Ace buried face down, and on average 1.6 of the four Aces start hidden. You cannot build what you cannot see. The column of seven is where the most cards are locked away.

3. A buried Ace is less of a disaster than it looks

Here is a result that surprised us. On deals where two or more Aces start face down, our player won 41.6%. On deals where no Ace is hidden at all, it won 44.5%. That is a difference of about three percentage points. A slow start is not a lost game, so keep going rather than restarting in frustration.

4. Use the stock, but don't lean on it

Our player's winning games went through the stock 2.3 times on average and took about 126 moves including draws. With unlimited passes there is no penalty for cycling, but every pass without a new tableau move is a sign the board, not the stock, needs attention. Look for a run you can split to free a card, or a column you can empty for a waiting King.

5. Don't panic at a frozen opening

In 8.6% of deals none of the seven face-up cards can move anywhere at the start. That feels bleak, but the stock is there for exactly this. Draw, and the position opens up more often than not.

6. Empty columns are for Kings, so plan the King

On this site only a King can fill a gap. Clearing a column with no King ready to take it wastes the space. Before you empty one, know which King (ideally one sitting on top of hidden cards) is going to move in.

How real players compare

The solitaire site solitairex.io published its own players' results: in the month before 16 July 2025, 913,106 games of classic Klondike were played there and 44.8% were won. That is remarkably close to our computer player's 42.3%. These are different groups playing different deals, so we treat it as an interesting coincidence rather than proof of anything. It does suggest that the average player wins a bit under half of their games, and that there is plenty of room above that.

The same leaderboard shows what dedication looks like: its top player had 7,765 wins at a 98.01% win ratio. You do not need to aim that high to enjoy a game, but it is a reminder of how much skill this "luck" game really involves.

Playing here

Klondike on free-solitaire.nz is free, and there is nothing to download, install or sign up for: no account and no email. It runs in any modern browser on a phone, tablet or computer, which makes it easy company on a slow evening or a rainy day at the bach.

• Large, clear cards and a zoom control for making the board bigger.
• Hint, unlimited Undo and Restart, so you can test the alternative lines our data says are usually there.
• Auto Play to send the cards home once the game is won.
• Drag and drop, or tap a card and then its destination.
• Wins, time and moves tracked for your current game.
• Sound you can switch off, fullscreen mode and a choice of table colours.

How we did the research, step by step

The free-solitaire.nz team ran this study in September 2026. We did not want a general figure for “solitaire”; we wanted to know how often the exact game on this page, with its own rules, can be won.

1. Reading the rules from this site’s own game code. We took the rules straight from the code that runs this page: seven columns of one to seven cards with the top card face up, a 24-card stock turned one card at a time with unlimited redeals, building down in alternating colours, any face-up run can move, and only a King may go into an empty column. The code also lets one card come back down from a foundation, so our solver allowed that too.
2. Rebuilding the game as a program. We wrote the game again in Python and checked that our computer player only ever made legal moves, over 2,000 test games (1,000 at Turn 1 and 1,000 at Turn 3).
3. Dealing the cards. Each deal comes from a numbered seed, so the same number always gives the same shuffle. We used deals 1 to 10,000. These are our own deal numbers, not the old Windows ones.
4. Choosing two kinds of player. The first is a bot that sees only what you see and never undoes a move. It follows a fixed order of priorities: foundation moves first, then moves that turn over a hidden card, then Kings to empty columns, and so on. Its result is a floor: a patient person with Undo should do better. The second is a solver that can see every card, including the face-down ones. It tries every legal move, up to 150,000 positions per deal. Its result is a ceiling no real player can reach.
5. Running it. The bot played all 10,000 deals and won 42.3%. The solver took deals 1 to 1,000: it found a win in 76.6%, proved 2.1% impossible and ran out of search on 21.3%. So between 76.6% and 97.9% of deals can be won with perfect knowledge.
6. Checking against published research. Blake & Gent, Journal of Artificial Intelligence Research, 2026 put Klondike Turn 1 at 90.480% with every card known. That sits inside our range, a good sign our model of the game is right.
7. Working out the margin of error. A “95% range” is the band the true figure falls inside 95 times out of 100 if you ran the same test again on fresh deals. For the bot, 42.3% came with a range of 41.3–43.3%.
8. What the numbers cannot tell you. The bot is simple, so 42.3% is not “how good people are”. And the solver left about one deal in five undecided, which is why we give a range rather than one number.

Every deal is seeded, so the whole test can be repeated exactly.

The technical side

The study is written in Python 3.11 using only the standard library. Klondike lives in its own module, klondike.py. A shared runner.py hands the deals out to 19 worker processes at once with Python's multiprocessing module, on a 20-core Windows machine. Deal number s is shuffled from a seed built from the game name and s, so the same number always gives the same cards.

• The bot follows a fixed priority list: foundation moves first, then any move that turns over a face-down card (starting with the column hiding the most), then Kings into empty columns, then waste cards, then a draw. It stops after a full pass through the stock with no progress. All 10,000 deals took about 14 seconds.
• The solver is a depth-first search over every legal move, including partial runs and cards coming back off the foundations. It keeps a set of positions already visited, so no position is searched twice, and gives up after 150,000 positions. Its only shortcuts are provably safe, which is why an exhausted search counts as proof. The 1,000 deals took about 17 minutes.
• The 95% range is a Wilson score interval, a standard way to put error bars on a percentage.

To check the bot, its moves were tested against the rules over 2,000 games before any figures were kept. An AI assistant (Claude, made by Anthropic) helped us write and review the code and draft this page, but it did not produce any figure: every number here comes from running the code, and we compared the results with published research before publishing.

By Admin, free-solitaire.nz. Study date: 18 September 2026.

Frequently asked questions

What percentage of Klondike games can be won?

With one-card draws and perfect knowledge of every hidden card, published research puts it at about 90.5%. Our own solver proved at least 76.6% of deals winnable. A computer player that plays blind, as you do, won 42.3%.

Is Turn 1 easier than Turn 3?

Much easier. Our computer player's win rate fell from 42.3% at Turn 1 to 10.6% at Turn 3, a roughly fourfold drop.

Should I restart when an Ace is buried?

No. Deals with two or more hidden Aces were won only about three points less often than deals with none. Play it out.

Is using Undo cheating?

Not here. There is nothing to win but the game itself, and Undo is the fastest way to find the better line that our data suggests usually exists.

Do I need an account?

No. Open the page and play. There is no sign-up and nothing to download.

What should I try next?

For a game where skill decides almost everything, try FreeCell, where 99.68% of our deals were solved. For a two-deck challenge, try Spider Solitaire. The full list is on the games page, and more of our findings are on the blog.