Press Start, watch a pattern of tiles light up and go dark, then click or tap the same tiles back from memory, in any order.
ReadytilesA sample pattern. Press Start, then tap the tiles that were lit.
More options
The standard run. Three tiles light for 1,500 ms on a 3 by 3 grid, and every cleared board adds one tile, up to 24 on a 7 by 7. A board forgives two wrong tiles and is lost on the third, and the third lost board ends the run. Your score is the biggest pattern you rebuilt.
–This run
–Your best
–Last run
Starts at 3 tiles and adds 1 after every cleared board, up to 24. The grid grows with the pattern, from 3 by 3 to 7 by 7, so fewer than half the tiles are ever lit. The pattern shows for about 1500 ms. Tap the tiles that were lit, in any order. Dark tile number 3 loses the board, which is a strike and a new pattern of the same size. The run ends at 3 strikes. The result is the biggest pattern you cleared.
3 by 3 grid: patterns of 3 and 4 tiles. Nine tiles, of which 3 or 4 light, so 84 or 126 possible patterns and 500 or 375 ms of flash for each lit tile. Small enough to take in as a single shape. Use these boards to settle where your eyes rest before the flash: on the centre tile.
No runs yet. Your last ten will show up here.
Today's boards
The same boards for everyone today. Finish a run and a badge counts your days in a row.
Five-second hold pattern climbThe pattern shows for 1,500 ms, then the grid stays dark and locked for 5 seconds before you may tap. Sperling's 1960 experiments with letter arrays found a detailed visual trace that fades within about a second of a display going dark. The hold outlasts it, so only what you stored as shapes is left.
Half-second pattern climbEvery pattern shows for 500 ms, a third of the usual flash. At 10 tiles that is 50 ms a tile, so there is no time to move your eyes round the grid: you get one look from wherever they were resting. It keeps its own best, because scores at the two flash lengths are not comparable.
One-miss pattern climbThe same climb with no forgiveness: the first wrong tile ends the run. It stops you sweeping the neighbours of a tile you half remember, and the score it gives is the pattern you can rebuild cleanly, not the one you can reach with two spare taps.
Show all 22 rounds and techniques (19 more)
Pattern climbs, continued
Pattern climb from 3 tilesThe standard run. Three tiles light for 1,500 ms on a 3 by 3 grid, and every cleared board adds one tile, up to 24 on a 7 by 7. A board forgives two wrong tiles and is lost on the third, and the third lost board ends the run. Your score is the biggest pattern you rebuilt.
Pattern climb from 8 tilesSkips the 3 by 3 and 4 by 4 grids and opens on 8 tiles of the 5 by 5, with the usual misses and strikes. Use it once the small grids have stopped costing you anything: reaching 13 tiles means clearing 6 boards here, where the full climb needs 11.
Sparse pattern climb on 7 by 7The grid is 7 by 7 from the first board, so 3 tiles sit among 49 and can form 18,424 patterns where the 3 by 3 grid allows 84. In 11,860 of them, about 64%, no two lit tiles touch, so there is often no shape to grab. This round trains exact placement: third row, second from the left.
Untimed pattern studyThe pattern stays lit until you press Hide, and nothing is scored or saved. Use it to practise carving a board into shapes out loud: bar, L, loose tile. When you can describe a 10-tile board in four words before you hide it, go back to a timed climb.
Mirror rounds from the research 2
Left-right mirror pattern climbEvery pattern is mirrored about the vertical centre line, so the left half matches the right, as in a face. It starts at 4 tiles and adds a mirrored pair after each cleared board. In a 2012 study that showed whole grid patterns at once, arrays with vertical symmetry, on the usual reading this left-right mirror, were recalled better than every other kind.
Up-down mirror pattern climbThe top half matches the bottom. By count this is exactly as easy as the left-right round: 8 mirrored tiles on the 5 by 5 grid allow 1,635 patterns either way, against 1,081,575 unmirrored. A 2018 study of matrix patterns found vertical and horizontal symmetry both raised scores, with no reliable difference between them. Compare your two bests to see which your eyes prefer.
Grid by grid 5
3 by 3 grid: patterns of 3 and 4 tilesNine tiles, of which 3 or 4 light, so 84 or 126 possible patterns and 500 or 375 ms of flash for each lit tile. Small enough to take in as a single shape. Use these boards to settle where your eyes rest before the flash: on the centre tile.
4 by 4 grid: patterns of 5 to 7 tilesSixteen tiles and 4,368 to 11,440 possible patterns. No more than 4 tiles fit on this grid without touching, corners included, so from 5 tiles at least two always touch and there is always a shape to grab. Two shapes and a loose tile cover most boards here.
5 by 5 grid: patterns of 8 to 12 tilesTwenty-five tiles, and 1,081,575 patterns at 8 tiles rising to 5,200,300 at 12. The flash per lit tile falls from about 188 ms to 125 ms. At 12 tiles some row must hold at least 3 lit tiles, so find the crowded row first and hang the rest of the picture from it.
6 by 6 grid: patterns of 13 to 17 tilesThirty-six tiles and between 2.3 billion and 8.6 billion possible patterns. At 15 tiles the flash works out at 100 ms a tile and at 17 it is 88 ms, so reading tile by tile is over. Split the grid into four 3 by 3 quarters and store one small picture for each.
7 by 7 grid: patterns of 18 to 24 tilesForty-nine tiles and from 11.6 trillion to 63.2 trillion possible patterns. At 24 lit, where the climb stops, 25 tiles stay dark, so the dark picture is no simpler than the lit one, and some row holds at least 4 lit tiles. Clearing 24 ends the run with a full score.
Techniques 8
Check yourself against the tiles-left countThe page counts the lit tiles you have still to find. If it reads 2 and the shape left in your head is an L of three, trust the count: one tile of the L is already tapped or was never lit. Tap the two you are surest of.
Count the dark rows and columns firstA fully dark row or column is the cheapest fact on the board. On the 5 by 5 grid, a dark top row and a dark right-hand column shrink the search to a 4 by 4 corner, which cuts 25 candidate tiles to 16 before you have stored a single shape.
Read shapes, not single tilesSeven lit tiles on the 4 by 4 grid might be a bar of three, an L and one loose tile: three things to hold. In Chase and Simon's 1973 chess studies, skilled players recalled briefly shown boards far better than novices, but much less so once the pieces were placed at random.
Rest your eyes on the centre tileAt 17 tiles a 1,500 ms flash gives 88 ms a tile, too short to visit each one. Before you press Start, look at the middle of the grid and keep your eyes there through the flash, so the whole pattern lands in one view and not in pieces.
Sit back until the grid fits one glanceOn a large monitor the 7 by 7 grid can be wide enough that its corners fall outside your sharp central vision, and you catch yourself scanning. Lean back or shrink the window until you can see all four corners while looking at the centre.
Stop dead after a wrong tileA board forgives two wrong tiles and is lost on the third. The usual second miss is a hurried tap on the neighbour of the first. Lift your hand, replay the shape against the edges of the grid, and only then tap again. One miss costs nothing; three cost a strike.
Store the simpler picture, lit or darkEleven lit tiles that fill the three left-hand columns of the 5 by 5 grid, apart from a dark 2 by 2 square, are hard as eleven tiles and easy as three columns minus a square. When the lit tiles nearly fill a region, remember the hole.
Tap your least certain shape firstOrder does not count here, so spend your freshest memory where it is weakest. If the board was a solid bar, a corner block and a vague diagonal of three, tap the diagonal at once and leave the bar, which you will still know in ten seconds, for last.
How this list was compiled: Rounds, grid notes and techniques written for this page. The counts were computed on 2026-09-19: the possible patterns of t lit tiles on a grid of n tiles are the binomial coefficient C(n, t) (C(9,3) 84, C(9,4) 126, C(16,5) 4,368, C(16,7) 11,440, C(25,8) 1,081,575, C(25,12) 5,200,300, C(36,13) 2,310,789,600, C(36,17) 8,597,496,600, C(49,3) 18,424, C(49,18) 11,554,258,485,616, C(49,24) 63,205,303,218,876); the grid rule is the smallest square from 3 by 3 to 7 by 7 on which the lit tiles are fewer than half, which gives 3 to 4, 5 to 7, 8 to 12, 13 to 17 and 18 to 24 tiles; the flash per lit tile is 1,500 ms or 500 ms divided by the tiles (1,500/3 = 500, 1,500/4 = 375, 1,500/8 = 187.5, 1,500/12 = 125, 1,500/15 = 100, 1,500/17 = 88.2, 500/10 = 50); the patterns of 8 tiles mirrored about one centre line of the 5 by 5 grid are C(5,0)C(10,4) + C(5,2)C(10,3) + C(5,4)C(10,2) = 1,635, the same for either line, and the formula was checked against a brute-force count on smaller grids; at most 4 tiles fit on a 4 by 4 grid with no two touching, corners included, because it splits into four 2 by 2 blocks; the fullest row holds at least the tiles divided by the rows, rounded up (12 on 5 rows 3, 24 on 7 rows 4); a run to 13 tiles takes 11 boards from 3 tiles and 6 from 8; and of the 18,424 patterns of 3 tiles on the 7 by 7 grid, 11,860 (64.4%) have no two tiles touching, corners included, counted by brute force. The research facts were checked on 2026-09-19. Rossi-Arnaud C, Pieroni L, Spataro P, Baddeley A, 'Working memory and individual differences in the encoding of vertical, horizontal and diagonal symmetry', Acta Psychologica 141(1): 122-132, 2012, https://doi.org/10.1016/j.actpsy.2012.06.007: the abstract (PubMed record PMID 22889674, read through https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=22889674&rettype=abstract) reports simultaneous presentation and that arrays with vertical symmetry were always recalled better than every other pattern. Hamilton CJ, Brown LA, Rossi-Arnaud C, 'Older Adults Benefit from Symmetry, but Not Semantic Availability, in Visual Working Memory', Frontiers in Psychology 8: 2373, 2018, https://doi.org/10.3389/fpsyg.2017.02373 (open access, read at https://pmc.ncbi.nlm.nih.gov/articles/PMC5787569/): vertical and horizontal symmetry both gave significantly higher spans than random matrix patterns, and the two did not differ reliably from each other (p = 0.114). Neither abstract nor the 2018 full text, nor the abstract of the 2006 paper that set the task (PubMed PMID 16338091), all re-read on 2026-09-19, says in words which axis 'vertical' means, and the figures could not be read. The term is read here in its standard sense of mirror symmetry about a vertical axis, the left half matching the right, and the copy says so with the words 'on the usual reading'; that this is the most salient mirror symmetry, as in a face, is from https://en.wikipedia.org/wiki/Symmetry. The account of Sperling's 1960 partial-report experiments (letter arrays shown for 50 ms, and a detailed trace that has decayed by about 1,000 ms after the display goes off) is from https://en.wikipedia.org/wiki/Iconic_memory. Chase and Simon's 1973 finding that skilled chess players recalled more of a briefly shown board than novices, and that the difference was much reduced when pieces were placed at random, is from https://en.wikipedia.org/wiki/Chunking_(psychology). No span, average or age figure from any of these studies is printed on the page. Licence: own work
How it works
This visual memory test lights a pattern of tiles for 1,500 ms, then turns the grid dark. Click or tap the tiles that were lit, in any order. Each cleared board adds a tile, from 3 on a 3 by 3 grid to 24 on a 7 by 7. Your score is the biggest pattern you rebuilt.
Read the full guide
How a board is won and lost
Press Start, watch three tiles light, and click or tap them once they go dark. On a keyboard, arrow keys move and Enter picks a tile. A dark tile is a miss. The third miss loses the board, which is a strike, and a new pattern of that size follows. The third strike ends the run. Under Round, One-miss pattern climb ends on the first wrong tile.
Why the mirror rounds are here
In a 2012 study, people who saw whole grid patterns at once recalled those with vertical symmetry best. On the usual reading, the left half mirrors the right. A 2018 study found horizontal symmetry helped too, with no reliable gap between the two. Play Left-right mirror pattern climb and Up-down mirror pattern climb, then compare your bests.
Read shapes, not tiles: a bar of three, an L and one loose tile are three things to hold, not seven.
At 15 tiles the 1,500 ms flash is 100 ms a tile, so look at the centre of the grid.
Order is free here (the sequence memory test is where it counts), so tap your least certain shape first.
Questions and answers
Why does the grid get bigger as I climb?
It grows so that fewer than half the tiles are lit. The tool picks the smallest grid that fits. Patterns of 3 to 4 tiles sit on 3 by 3, and 5 to 7 on 4 by 4. Then 8 to 12 use 5 by 5, 13 to 17 use 6 by 6, and 18 to 24 use 7 by 7. Phones follow the same rule.
Why is the five-second hold so much harder?
It removes the fading picture. Sperling's 1960 experiments with letter arrays found a detailed visual trace that decays within about a second of a display going dark. Your first taps in the standard climb can lean on that trace. After the wait in Five-second hold pattern climb, only the shapes you stored remain.
Can this test show whether my memory is getting worse?
No. A lower score usually means a tired evening, a smaller screen or a finger where you used a mouse before. This is a game, not a cognitive assessment, and it cannot detect, track or rule out any condition. If your memory worries you, ask a doctor.