How to Win at Solitaire: Klondike and Spider Strategy

A ranked order for your next move in Klondike, what an empty column is really worth in Spider, and an honest look at how many deals can be won at all.

Published August 23, 2026 · By Shivam Kumar · 11 min read · Strategy

☰ Table of contents 8 sections

1. How Many Deals Can Be Won at All
2. The Safety Net Is Different in Every Game Here
3. Klondike: How to Rank Your Next Move
4. Spider: What an Empty Column Is Worth
5. When to Give Up on a Deal
6. Further Reading
7. FAQ
8. Play One Deal With One Rule

PlayZone has three games from the solitaire family — Solitaire, which is Klondike, plus Spider Solitaire and Mahjong Solitaire — all free in the browser, no download and no account, and I built all three myself. The rules are on each game's page; this is about what they never answer. When you lose, was the deal winnable at all, and how much of the loss was you?

I have not run a solver and I have not tracked my own win rate. Every statistic below is somebody else's published work, cited where it appears; every mechanical claim is read out of the game source files. Where the code disagrees with something already published on this site, I say so.

How Many Deals Can Be Won at All

Start with Klondike, which people assume is mostly luck. The best current estimate comes from Charlie Blake and Ian Gent, whose solver Solvitaire puts three-card-draw Klondike at 81.945% ± 0.084% (Source: Blake and Gent, Journal of Artificial Intelligence Research, 2026). That figure carries a large asterisk: it is for the "thoughtful" variant, where the player already knows every face-down card. It describes the deal, not you.

Draw size is one of the big rule levers: thoughtful winnability runs 90.480% ± 0.116% at draw one and 69.337% ± 0.098% at draw four. PlayZone's Klondike draws one card at a time with unlimited recycling, so roughly 90% is the ceiling here — for somebody who can see through the board.

Underneath sits a floor of a different kind. Sampling ten million deals, an earlier study proved 855,741 of them dead, unwinnable however they are played, putting at least 8.56% of draw-three deals beyond rescue (Source: Bjarnason, Tadepalli and Fern, ICGA Journal, 2007). For the draw-one game here the newer work gives a tighter floor: a 90.480% ceiling means roughly one deal in ten is dead whatever you do.

For Klondike as ordinarily played, cards hidden, the strongest established result is that some strategy wins over 35% of deals (Source: Bjarnason, Fern and Tadepalli, ICAPS 2009) — a lower bound on good play, not an estimate of it. The honest answer for draw three is a band between about 35% and 82%. Researchers have called this "one of the embarrassments of applied mathematics that we cannot determine the odds of winning the common game of solitaire" (Source: Yan, Diaconis, Rusmevichientong and Van Roy, NIPS 2005).

Spider is the opposite problem

Four-suit Spider, the version everybody calls brutal, is almost never limited by the deal. Alex Robinson's plspider project has now played more than 131,000 four-suit deals: 65,000 under PySolFC's numbering with 64,998 won, and 65,365 under WinSpidr's with 65,364 won. The three still outstanding are listed as "not yet won" rather than proved unwinnable, and Robinson's own estimate is one unsolved deal in 40,000 (Source: Robinson, Winnable Spider Solitaire Games).

Blake and Gent report 98.487% ± 1.513% for the same game, which looks like a contradiction and is not one: of their ten thousand Spider instances, 9,731 were proved winnable, 269 ran out of time, and none was proved unwinnable — the wide interval is the timeouts counted both ways. I could not find a published study of one-suit or two-suit Spider, and the percentages circulating for those trace back to nothing I could verify.

That inversion is why this post exists. In Klondike the deal is the constraint; in Spider it almost never is, because the difficulty there is hidden information and committed moves. A lost Spider game is far more likely to have been winnable than a lost Klondike one.

The Safety Net Is Different in Every Game Here

How much caution a move deserves depends on what the software will do for you, and the three differ. Mahjong Solitaire generates its deals solvable by construction: the code simulates removing random free pairs from the full 72-tile layout before any faces are assigned, then paints the faces on in that removal order. There is one caveat worth knowing. That construction is retried up to sixty times, and on the rare occasion all sixty get stuck the generator falls back to randomly shuffled faces, which may not be fully solvable — so “generated solvable” is very nearly, not quite, a guarantee. Hint and Shuffle sit there anyway, because solvable at the deal does not mean every line wins.

Klondike here is the opposite extreme. The deck is a plain shuffle with no solvability filter, and there is no undo, no hint, no auto-finish and no stuck detection anywhere in the file. The only control on the page is New Game. A committed move is committed.

Spider sits between them. Its deal is unfiltered too, but the code snapshots the game before every move and every stock deal, and undo restores the board, the stock, the cleared count, the score and the move count — including handing back a sequence and its 100 points. One limit matters: the history is capped at sixty snapshots and the oldest is dropped silently, so sixty moves is as deep as undo goes.

Klondike: How to Rank Your Next Move

Having several legal moves is the normal state of a Klondike board. The guide already puts exposing face-down cards first and already gives the tie-break between two equal flips, so neither is mine to claim. What it does not say is how the rest trade against each other when they compete. Here is the order I use, strongest first.

1. Take any move that turns a face-down card face up

Information is the currency: the seven columns hide twenty-one cards at the deal, and every one you expose narrows what can go wrong, so a flip beats a tidy-up. If two flips compete, the guide's tie-break holds: dig the column with more face-down cards beneath it, because long columns are where the game hides its problems.

2. Send cards up late, but remember you can fetch them back

The usual warning against rushing cards up is right: a 4 on a foundation cannot catch a 3 in the tableau. What the guide does not mention is that this build lets you take it back — tap the top card of a foundation, then a tableau column, and it comes down if it fits. The move has a formal name, worrying back, and Blake and Gent measured how often it is indispensable: 0.52% ± 0.04% of winnable draw-three deals cannot be won without it, and 0.34% ± 0.08% at draw one, the rule this build uses. An Ace can only return onto a 2 of the opposite colour, so an early Ace is usually stranded.

3. Spend the stock freely, because it is not the scarce resource

This Klondike draws one card per tap, and when the stock empties a tap recycles the whole waste pile, with no limit on passes. The recycle turns the waste back over, so the stock returns in its original order: every pass presents the same sequence, and all twenty-four cards stay reachable. Blake and Gent rely on the same fact: drawing by one with unlimited redeals makes every stock card available at any time. The only cost is moves and time: a draw and a recycle each increment the counter, and the clock starts on your first move, pausing while the tab is hidden.

4. Decide which King fills the space before you make it

An empty column takes a King, or a King-led run, and nothing else. In the same paper's rule experiments, letting any card fill a space instead of only a King lifts draw-three winnability from 81.945% to 94.959% ± 0.045%, and forbidding spaces altogether collapses it to 2.168% ± 0.121%. So check which King you have, and in which colour, before you empty a column: a space you cannot fill is worth nothing, and the wrong-colour King blocks the Queen you needed. One detail makes tidiness load-bearing: the run check reads from the card you tap to the bottom of the column, so a neat run with a stray card beneath it will not lift as a unit.

Spider: What an Empty Column Is Worth

Spider inverts that rule. An empty column accepts anything — any card, any group, any suit. It is not a slot waiting for a King but a universal parking space, and the most valuable thing on the board.

The space you are taxed for holding

The catch is in the stock: the code refuses to deal a row while any column is empty. An empty column is therefore a tool you can hold or spend, never both. The price of spending it is misleading: a deal costs one point and one move, the same as sliding a single card, but it drops a face-up card on all ten columns, and most of them bury something.

What an off-suit join actually costs

A card lands on any card one rank higher regardless of suit, but a group only lifts if it is a face-up same-suit run descending by one, checked from the card you tap to the bottom of the column. An off-suit join is therefore a loan you repay one card at a time: take them when they buy a flip or a space, not when they are merely tidy. Completed King-to-Ace runs vanish the instant they form and flip the card underneath, so finishing one doubles as excavation. Two-suit play uses spades and hearts, one black and one red, so colour alone tells you whether a join is same-suit.

What undo refunds, and where it stops

Because undo restores the score and the move count exactly, exploring a line and reversing it costs nothing but the sixty-snapshot ceiling. Play a candidate out four or five moves, see what it did to the face-down cards, then undo back to the decision point. Two edges: undo switches off the moment the eighth sequence lands, and the D key deals a row with no confirmation and no mention in the guide.

When to Give Up on a Deal

Neither game will ever tell you a deal is dead. Klondike only checks for fifty-two cards on the foundations, Spider only for eight completed sequences; neither file has stuck detection. The judgement is yours — a deliberate contrast with the layer method for the Slide Puzzle, where the board is scrambled by legal moves and can always be walked back. There, if you are stuck, the fault is in your route; here it might not be.

My rule of thumb in Klondike: if I have cycled the whole stock twice without exposing a face-down card and without a new tableau join appearing, the deal is finished. That is not a proof; nothing here can give me one. What the research should change is not your technique but your reading of the result. Losing a Klondike deal is weak evidence that you played badly; losing a Spider deal is much stronger evidence, and that asymmetry is the most useful thing I know about these two games.

Further Reading

• Play Slide Puzzle Online — the other side of the coin: every board has an answer.
• How to Beat 2048 — the same discipline elsewhere: pick an anchor, stop reacting.
• Pelmanism, the classic concentration game — the nearest card-game relative here.
• Quick games for a five-minute break — where a deal fits in a short break.

FAQ

Is every solitaire deal winnable?

No, and the three differ. Mahjong Solitaire builds its deals so a solution exists before you start, bar a fallback its own code calls extremely unlikely. Klondike and Spider are plain shuffles with no filter. The risk is real in Klondike: for the draw-one rules this build uses, solver work puts thoughtful winnability at 90.480%, so roughly one deal in ten is dead whatever you do. In four-suit Spider it is close to nonexistent — no study I could find has ever proved one deal unwinnable.

What are the real odds of winning Klondike?

There is no settled figure. The published percentages — about 82% at draw three, 90% at draw one — describe "thoughtful" play, where the solver knows every face-down card. For hidden-card play the honest answer is a band: above 35% at the floor, about 82% at the ceiling.

Can I take a card back off a foundation?

Yes, though the game's own guide does not mention it. Tap the top card of a foundation, then a tableau column, and it returns if it fits by rank and colour. It is rarer than you would hope: in Blake and Gent's runs, 0.34% ± 0.08% of winnable draw-one deals cannot be won without it.

Is Spider's undo really unlimited?

Not quite. The history holds sixty snapshots and silently drops the oldest, so sixty moves is as far back as you can step, and undo switches off once you have won. It does restore the score and move count exactly, so reversing a move costs nothing.

Is four-suit Spider hard because of the deal or because of me?

Because of you, almost certainly. Published work on four-suit Spider has found all but a handful of deals winnable with full information and never proved one unwinnable: the difficulty is hidden cards and irreversible commitments, not the shuffle. Be wary of per-difficulty claims, though — I could not find a published study of one-suit or two-suit Spider.

Do these games save my deal if I close the tab?

No. None of them saves a game in progress, so reloading deals fresh and Spider reverts to one suit. Your device keeps Klondike's fastest winning time, Mahjong Solitaire's fastest clear time, and Spider's best score for each suit count separately — all written only when you win.

Play One Deal With One Rule

Open Solitaire and play one deal under one restriction: before every move, ask whether it turns a face-down card over, and if one does, take it. It will feel slow, and it will win deals you would otherwise lose. Then open Spider Solitaire on one suit and refuse to touch the stock until you can honestly find no move on the board. Those two habits are most of the gap between losing to the deal and losing to yourself.