8 ways to improve your short-term memory, from chunking and the memory palace to sleep and exercise, plus what brain training really does. With memory tests.
Published November 2, 2026
To improve your short-term memory, change what you put into it, not how big it is. Your short-term memory holds about four chunks at a time, and that limit barely moves with practice. But a chunk can be one digit or a whole year, one tile or a whole shape. People who remember 20, 40 or 80 items have not grown a bigger memory: they have learned to pack more into each chunk.
Here are 8 ways to remember more, each with the research behind it, and an honest look at brain training. Take the HumanRank number memory test first, so you have a baseline to beat.
Remember the number, type it back, and see how many digits you can hold before you try the tips.
Take the number memory test
Short-term memory is what holds a phone number between reading it and dialing it. It keeps a small amount of information available for a few seconds, and it loses it fast: in a classic 1959 experiment by Lloyd and Margaret Peterson, people who were kept from rehearsing three letters recalled them only about 10% of the time after 18 seconds.
Psychologists often say working memory instead. The difference is what you do with the information: short-term memory stores it, working memory stores it and works on it, like adding two numbers in your head or reversing a sequence. Every memory test on HumanRank leans on both.
George Miller's famous seven, plus or minus two was the starting point. In 2001, Nelson Cowan reviewed the evidence and put the real limit nearer four chunks once you take grouping away. Seven only looks possible because we chunk without noticing.
Medians of the Human Benchmark score distributions, between 22,000 and 36,000 results per test, converted to the number of items remembered. Verbal memory spreads over a whole run of words, so it also tests longer-term recognition.
All of these sit above Cowan's four chunks, and that is the point. Eight random digits are two or three chunks for most people, and twelve tiles are a few shapes. The tips below are ways to make the chunks bigger. For the full percentiles of each test, see the number memory, sequence memory, chimp test, visual memory and verbal memory averages.
Group items into units you already know. 1-9-8-4 is one year, not four digits, and 5820 4917 is two chunks, not eight digits. Chunking is the single biggest lever on this list: one student trained by K. Anders Ericsson and William Chase went from 7 to 79 digits in about 230 hours of practice, by hearing digits as running times. His memory span never changed; his chunks did. The full story is in our number memory test article.
Repeating something under your breath keeps it alive past the few seconds it would otherwise last. Your inner voice has a limited loop, though. In 1975, Alan Baddeley, Neil Thomson and Mary Buchanan found that people remember fewer long words than short ones, roughly as many as they can say in about two seconds. So rehearse fast and short: read digits as pairs ("fifty-eight, twenty") instead of one by one, and keep a steady rhythm, like reading out a phone number.
The method of loci, or memory palace, places each item along a route you know well, like the rooms of your home, and walks the route to recall them. It is the main technique of memory champions. In 2003, Eleanor Maguire and colleagues scanned 10 of them and found no difference in brain structure from ordinary people; most simply used this method.
It can be learned. In a 2017 study by Martin Dresler and colleagues, people with no training practiced the method of loci for 30 minutes a day over 6 weeks. Their recall of a 72-word list rose from 26 to 62 words, while a group that trained on a working memory game barely improved, and the gain was still there four months later. A memory palace takes a few seconds per item, so it helps most when you get time to study, as with long numbers or word lists.
On visual tests, the same chunking works with shapes. In the visual memory test, see the lit tiles as an L, a diagonal or a letter instead of twelve separate squares. In sequence memory, trace the path as one movement, like a gesture drawn on the board. In the chimp test, memorize the route from 1 to the last number before the digits disappear, not each position.
The Peterson experiment looked like proof that short-term memory simply fades. Then, in 1962, Geoffrey Keppel and Benton Underwood looked at the first trial of each session and found almost no forgetting at all. What wiped memory out was the earlier trials crowding in. Short-term memory is not so much leaky as easily overwritten. While you hold something, stop taking new information in: look away from other numbers, don't read the next item early, and wait to talk until you have used what you are holding.
Tiredness hits memory before you notice it. A 2010 meta-analysis by Julian Lim and David Dinges pooled 70 studies of short-term sleep deprivation: the largest effects were on attention, and working memory dropped clearly too. A night of short sleep will not erase what you know, but it lowers how much you can hold and how long you stay focused on it. Test your memory after a full night, not at 2 am.
Aerobic exercise is one of the few things shown to change the brain itself. In a 2011 trial led by Kirk Erickson, 120 older adults either walked for 40 minutes three days a week or did stretching for a year. The walkers' hippocampus, a region central to memory, grew by about 2%, while it kept shrinking in the stretching group, and the walkers' spatial memory improved. The effect builds over months, not before a single test.
Practice works, but it stays where you practice it. Playing number memory every day will make you better at number memory, mostly because you build better chunking habits for it. It will not make you remember names or groceries better. If you want a better grocery memory, practice remembering groceries.
Not beyond the games themselves. In 2010, Adrian Owen and colleagues ran a 6-week online brain training study with 11,430 participants for the BBC. Players improved on every task they trained, but showed no gains on untrained tests, even ones that were closely related.
Working memory training tells the same story. Monica Melby-Lervåg and Charles Hulme reviewed 23 studies in 2013, then 87 publications in 2016 with Thomas Redick. Training reliably improved the trained task and very similar ones, but did not carry over to reading, arithmetic or intelligence. The dual n-back game made headlines in 2008 for raising fluid intelligence, and later replications did not find the effect.
So treat memory tests as a measure, not a cure. They show you where you stand, let you see whether chunking or a memory palace helps, and are fun to beat. Memory also changes with age: working memory peaks between about 25 and 30, as we cover in at what age does your brain peak.
See the tiles as shapes, not squares, and find out how far chunking takes you.
Try the visual memory test
Chunk what you need to remember into bigger units, like reading digits as pairs or years, and rehearse it in your inner voice without taking in anything new. Both work immediately. Sleep and regular exercise help over weeks and months.
About four chunks, according to a 2001 review by Nelson Cowan. George Miller's famous figure of seven, plus or minus two, counted items that people had quietly grouped together. With bigger chunks, the same four slots can hold far more.
On Human Benchmark's memory tests, the typical player remembers about 8 digits on number memory, 9 tiles in order on sequence memory and 12 tiles at once on visual memory.
They improve your score on the games you play and on very similar tasks. Large studies, including one with 11,430 brain trainers in 2010, found no gain on untrained memory tasks or on intelligence.
Short-term memory holds information for a few seconds. Working memory holds it and lets you work on it, like doing mental arithmetic or putting a list in reverse order. The two overlap, and most memory tests measure both.
Yes. In a 2017 study, people with no training practiced the method of loci 30 minutes a day for 6 weeks and went from recalling 26 to 62 words out of 72. The gain was still there four months later.
