Memorising Chinese Characters

To memorise Chinese characters, learn the components they are built from first, then give each new character a short story in which its components act out its meaning, add a separate cue for its sound, and recall it with the character covered the next day and at widening gaps after that. Take 好, good, which is 女, a woman, beside 子, a child: picture a woman lifting her child and thinking how good this is. Its reading, hǎo, sounds like English how and carries the third tone, the low dipping one, so her "how?" dips too. Once 女 and 子 are known, that one picture is the whole job.

How a character is built

Most characters are not pictures but compounds of smaller components, many of them characters in their own right, squeezed to fit their place in the block. A component may carry meaning or sound, and the commonest kind of character has one of each: the phono-semantic, or semantic-phonetic, compound. Of the 3,500 characters used most often in Standard Chinese, about 58 percent are phono-semantic compounds, while characters built only from meaning, the kind mnemonic books love, are the rarest at about 5 percent. Across all the characters of modern written Chinese, over 90 percent began as phono-semantic compounds, but sound and meaning have shifted since, and in many of them the components no longer do their original job.

河, river, shows the pattern. The three strokes on its left, 氵, are the combining form of 水 (shuǐ), water, and say what the character is about, while the right side, 可 (kě), says roughly how it sounds, and 河 is read hé. Meaning components tend to sit on the left or the top and sound components on the right or the bottom, though the rule is loose and the exceptions are many.

Sooner or later you will meet the word radical. Radicals are the components a dictionary files characters under: the standard 214, first listed in the Zihui dictionary of 1615 and taken up by the Kangxi Dictionary of 1716, or the 201 that China has promoted as the national standard for simplified characters since 2009. A character has exactly one radical, which is often its meaning component and sometimes not. John Renfroe of Outlier Linguistics, writing on Hacking Chinese, calls radicals "an imperfect, man-made system of arranging and looking up characters in a dictionary" rather than the building blocks, and the useful habit follows from that: learn the components that recur, whatever list they appear on, and learn what each one is doing in the character in front of you.

Learn the components first

The case for components first is arithmetic. A story that turns a component into something arbitrary helps you with one character, while a component you recognise helps with every character that contains it. Olle Linge of Hacking Chinese makes the point with 口, mouth: draw it as an ice cream to remember 吃 (chī), to eat, and you are stuck with ice creams in 吗 and in 喝 (hē), to drink, whereas "mouth" makes sense of most characters that contain it. As he puts it, "Your goal is not to learn one character after all, but to learn many."

Remembering Simplified Hanzi, James Heisig and Timothy Richardson's adaptation of Heisig's kanji course, is built on this. Its first book covers the writing and meaning of the 1,000 most common simplified characters plus 500 more best learned early, and the second adds another 1,500 for 3,000 in all. The order is the method: simple components nest inside simple characters, which then serve as parts of harder ones, so most new characters arrive with their pieces already known. You can borrow the principle without the book. Learn a component when it first turns up in a character you want, give it one fixed picture and keep it, and if a component looks rare, do as Linge suggests and wait until it reappears in a second character before investing in it.

These eight components carry the worked examples below, each checked against Wiktionary and, for the forms on the left, Wikipedia's account of radicals.

One story per character

Heisig's book calls its technique "imaginative memory": each character gets a key word for its meaning, and a short story ties its components, which it calls primitive elements, to that key word, so that the parts, and with them the writing, come back with the meaning. Two habits separate a story that lasts from one that misleads.

Use the real story when there is one. Many characters still make sense: 休, to rest, is a person, 亻, beside a tree, 木, and Wiktionary's account of its origin is the story itself: "a man leaning against a tree, resting." Look a character up before you invent anything. Linge points readers to the Outlier dictionary and Zi.tools for this, and Wiktionary's free entries supplied every origin in this guide; where the true story is lost or convoluted, invent freely.

Give every working component a part to play. Linge's rule is that "your mnemonic should have the same number of functional components as the character does", a functional component being one that does a job in the character, such as giving its sound or meaning. A story that leaves a component out, or merges two into one picture, leaves you nothing to write it from.

Make the story an action, as in any link method image, and keep it to a single moment rather than a plot. The test runs both ways: thinking of rest should give you the person and the tree in their places, and seeing 休 should give you rest.

Meaning, sound and shape are three jobs

A character holds more than its meaning. Linge lists what is encoded in one, from the meaning and the pronunciation (initial, final and tone) to what the components mean, how they are written and where they sit, and he warns that a system to memorise all of it "would be extremely time-consuming and ultimately unnecessary." His rule is to use a mnemonic only for the part you keep losing: "If you keep forgetting the tone, create a mnemonic for the tone," and if you do not forget it, leave it alone.

So treat the three as separate jobs and test them separately. The component story carries meaning and shape, and the sound needs a cue of its own, because a story about a woman and a child says nothing about hǎo. Meaning and shape come first, from the story; the sound comes second, from the phonetic component where there is one and a keyword where there is not; and the tone comes last, by whichever of the methods below you find you need.

Sound: read the phonetic component first

Because most characters are phono-semantic compounds, the first sound cue is usually inside the character already. 马, horse, is read mǎ, and it lends that sound to a whole family:

Linge gives another family: 碟, 蝶, 牒 and 喋 have different meanings but share a phonetic component, and all are read dié. The fit is often looser than that. The tone may differ (样 yàng against 羊 yáng), or the initial (很 hěn against 艮 gěn), or the final (狼 láng against 良 liáng), and sound change has erased some clues altogether: 偷 (tōu), to steal, and 输 (shū), to lose, share the phonetic 俞 (yú), yet sound nothing alike. Even a loose fit beats a bare memory, so learn what the phonetic gives you and save the mnemonic for the difference.

When a character offers no clue, use the keyword method: an English word that sounds like the syllable, built into the story. Hǎo is how, so the woman holding her child asks how anything could be better. The keyword covers the consonant and vowel, and the tone, as always, is a separate cue.

Tones: the honest difficulty

Mandarin has four main tones and a neutral one on weak syllables, and the tone is part of the word: mā is mum, mǎ a horse, mà (骂) to scold and an unstressed ma the question particle. There are only about 400 different syllables when tone is ignored and about 1,300 when it is counted, so the language leans on tone to keep words apart, and Wikipedia's account of its sound system notes that many non-native speakers "have difficulties mastering the tones of each character."

Mnemonics help less here than anywhere else, and Linge says why: in conversation "you have only a split second to produce the right tone," and no story is that fast. His defence of them is that slow recall still beats guessing, which "might reinforce the wrong tone," and that a mnemonic is a steppingstone, never all you have if you are also reading and listening to a lot of Chinese. It holds the tone until that exposure takes over. There are three ways to put the tone into the picture, from lightest to heaviest.

A motion per tone. The keyword guide gives each tone a fixed action inside the image: the first floats level, the second lifts, the third dips and bounces back, and the fourth slams down. It costs nothing to learn and suits a learner with a few hundred characters.

Actors for the sounds and places for the tones. Actor-and-location systems encode the whole syllable. In Mandarin Blueprint's Hanzi Movie Method, people from your life stand for the initials, buildings you know well for the finals, and rooms for the tones: "Outside the entrance represents the first tone, in the hallway or kitchen, represent the second tone, bedrooms and living room represent the third tone, the bathroom and backyard represent the fourth tone, and the fifth tone is on the roof." The Marilyn method, which Serge Gorodish published in 2012 and the Mandarinbanana site documents, is built the same way. "Marilyn initials are represented by personas, while Marilyn finals are represented by locations," and the tone is the spot within the location, from in front of it for the first tone through the kitchen, the living room and the bathroom to the roof for the fifth. Its example casts Don Quixote as the initial d, and the site counts 514 characters whose pronunciation starts with d, every one of which can share him.

These systems are a memory palace for sounds, and they cost what a palace costs: a cast and a set of buildings learned in advance. Mandarinbanana says plainly that "you'll need more than a day to learn the Marilyn method", and Mandarin Blueprint says it "takes between 5 and 20 characters to get the hang of it". Both sites vouch for the systems they teach, and Mandarin Blueprint promises you will read, pronounce and write any character "in less than one minute, and never forget"; we found no independent test of either.

A cue only where you fail. This is Linge's rule applied to tones: no system at all, and a tone cue added only to the characters whose tone you keep getting wrong. A sensible default is the motion per tone from the first day, adding cues to the characters you keep mispronouncing, with a full actor-and-location system kept for the learner aiming at thousands of characters who enjoys the bookkeeping.

Four characters, worked

Each of these was checked against its Wiktionary entry for components, meaning and reading. Read each scene slowly enough to see it, then cover it and test yourself in both directions.

好 hǎo, good. Wiktionary analyses it as 女, woman, plus 子, child, and gives as the widely accepted reading that it was good for a woman to have a child, while noting other theories. Picture a woman on the left lifting a child on the right and beaming. There is no phonetic component, so the sound comes from the keyword how, and the third tone from her voice: "how could this be better?", dipping low on how. From good you should be able to write woman, then child; from 好 you should see the scene and say hǎo.

妈 mā, mum. This is a phono-semantic compound, with 女, woman, for the meaning and 马, horse, for the sound. Linge says a picture of a horse riding on your mother's back is fine "as long as you know that 马 is actually a sound component." The sound comes free, since 马 is mǎ, and only the tone needs a cue. 妈 is first tone, so the horse rides perfectly level on her back, where 马 on its own dips. Learn 妈 and its family together: 吗, a mouth plus the horse, is the question particle ma, said lightly at the end of a question.

休 xiū, to rest. Here 亻, the form 人 takes on the left, stands beside 木, a tree, and the real story is the mnemonic: "a man leaning against a tree, resting." The sound has no clue in the character, which is where an actor-and-location system earns its keep. Your actor for the initial x leans against a tree outside the entrance of the building you use for the final iu, outside the entrance because xiū is first tone. Without such a system, find a keyword of your own for xiū and let the resting figure float level.

想 xiǎng, to think. This phono-semantic compound breaks the placement habit, with the sound component 相 (xiāng) on top and the meaning component 心, heart and mind, underneath. 相 is itself a tree, 木, beside an eye, 目, so the scene is an eye peering out from behind a tree that has taken root in a heart, thinking. The reading comes from 相, and only the tone changes, from first to third, so the thought dips. Linge uses the same character to make a point worth keeping: a beginner's mnemonic for 想 "obviously fades away over time," because after meeting it hundreds of times you no longer need it.

When mnemonics backfire

Linge's April 2026 article on bad mnemonics names three failures, and each one is a habit to drop.

Ignoring the real story when it helps. The ice cream standing in for 口 in 吃 is not wrong for 吃 alone, but it throws away what Linge calls "an extremely common character component." "So do look at the origin before you make stuff up," he advises.

Hiding the structure. A Chineasy picture draws 猪, pig, as a single pig. In fact it is 犭, a component marking animals, plus 者 (zhě), which gives the sound, and the picture hides both, so it is no help when 猫 (māo), cat, turns up with the same 犭. This is the failure the functional-components rule above prevents.

A shape that does not match. Two children's pictures for the Zhuyin symbol 冂 show a little sister and a cat, each linked to it only by sound. "The shape is just arbitrarily written on top of the drawing; the two have nothing in common," so neither the picture nor the shape can call up the other. Linge's fix is 门 (mén), door, which looks like 冂 and starts with the same m.

A popular story is not always the true one. 明, bright, is often taught as sun beside moon, but it is probably a simplified form of an attested variant, 朙, which can be read as a phono-semantic compound, and 秋, autumn, often explained as grain plus fire, became a phono-semantic compound in its later form. A false story can still be a good mnemonic, as the horse on mother's back is, provided you know which component does which job.

Spaced review, both directions, and a pen

A story gets a character in; review keeps it. Recall each new character once within the hour, as part of building it; the first spaced review is the next day, and to keep the set for good, use gaps of about 1, 6 and 16 days, then monthly. The spaced repetition guide explains why the first review is tomorrow rather than tonight and why the gaps grow.

Review in two directions, because they fail separately. Reading means seeing the character and producing its meaning and sound, and writing means seeing the meaning and producing the character, by hand. The pen matters. Connie Qun Guan, Charles Perfetti and colleagues taught adult learners of Chinese with an online tutor and found that writing characters by hand led to better word recognition and stronger links between character and meaning than reading alone, though not to a better grip on sound, while typing pinyin strengthened only the sound. Write the character when you review its shape, and say the reading aloud when you review its sound.

Expect the stories to fade, and let them. After enough clean recalls the meaning arrives before the scene does, which is the mnemonic finishing its job, and the scene is still there to rebuild a character that later goes dark.

How many characters, and how fast

The current HSK syllabus, issued in November 2025 by the Centre for Language Education and Cooperation, which sits under China's Ministry of Education, and dated on its cover for use from July 2026, sets separate lists for words, for characters to recognise and for characters to write. Counted from its lists, cumulatively:

Two things stand out. Even at the top the syllabus asks you to write fewer than half the characters it asks you to read, and the jump from level 6 to levels 7 to 9 adds more characters than the first four levels together. Study lists made for the previous six-level test, introduced in 2010, use different numbers, and as of July 2026 exams on the new syllabus had not fully replaced it, so check which version your exam and your list follow before you plan around them.

As for pace, no study sets a rate for characters. We suggest ten new ones a day, the keyword guide's figure for words, because that is about what survives the review load. At ten a day the 246 characters of level 1 take under a month, the 1,096 of level 4 about four months, and the 1,500 of Heisig's first book about five, which is our arithmetic rather than anyone's measurement. Hold the rate down when reviews start to take longer than the new characters do.

A first week, day by day

By the end of day 8 you have about eighty characters and a routine that keeps them. From day 9 each day brings one six-day recall, of the batch built a week earlier, and from day 24 one sixteen-day recall as well. Start every session with the reviews, and whenever a character fails, ask which job failed: a missing meaning means the story was weak, a wrong component means the story left one out, and a wrong tone wants a cue. Rebuild that part rather than rereading the whole card.

Doing it on this site

This site has no Chinese decks of its own, so the deck is yours to build, which is as it should be for stories you made yourself. Save your characters as a .csv file with a header row such as front,back,note: the character on the front, its meaning and reading on the back, and your story in the note, which the importer joins to the back. Upload it at the import page (it also takes .tsv, Anki's .txt export and .apkg files), where you confirm which column is the front and which the back, and it creates a deck named after the file in My Dict.

The custom drill then runs it. Pick the deck, five, ten or twenty cards, and a memorise window of fifteen seconds to three minutes, during which the cards are shown one at a time with their backs; the recall that follows covers only the cards you saw, so let the timer run, or step through every card with "next", before you press "ready to recall". Each character then appears alone: say its meaning, reading and story, press "show note" and grade yourself miss, hard or easy, and a missed card comes back at the end of the session. Cards that are due come first, most overdue first, then new or edited ones and then the rest, in the deck's order; it never shuffles. The drill tests reading, because the character is the prompt. For writing, save a second file with the columns swapped, meaning and reading on the front and the character on the back, and write each character on paper before you reveal it.

What the research says

The research on mnemonics for characters is small and mixed: a story helps a character in, but whether it helps the character stay differs from study to study.

Wang and Thomas ran two studies in 1992 comparing imagery mnemonics with rote learning for the English meanings of Chinese characters, testing immediately and after two days or a week, and "in no instance was there any indication that imagery-based mnemonics conferred an advantage beyond the immediate test of recall." The mnemonic groups forgot more, and the authors read the result in terms of the different long-term effects of experimenter-imposed and self-generated encodings.

Helen Shen in 2004 compared rote memorisation with two kinds of elaboration, learners' own and instructor-guided, and found that deeper processing gave significantly better retention of both sound and meaning than rote. The instructor-guided version beat the learners' own at a 20-minute interval, but "this advantage disappears at a 48-hour recall interval." In Kuo and Hooper's 2004 study of 92 high-school students, those who generated their own mnemonics outscored the groups given visual or verbal mnemonics and a translation-only group, and also spent the most time on task. Guan and Perfetti's writing studies, above, add that a pen helps the shape and meaning but not the sound.

Put together, a story helps a character in, a story you make yourself does at least as well as a borrowed one once a day or two has passed, and a mnemonic's lead over rote learning can vanish within days, the same pattern the keyword method shows for vocabulary and the reason the reviews come first. None of these studies tested Heisig's course or an actor-and-location system over months, and for those we found only the practitioners' word.

Tonight

Learn 人, 女, 子 and 木, then 好 and 休 with their stories and sound cues, and cover them once within the hour. Tomorrow morning, before you look at anything, write both from their meanings and say their readings aloud. If both come back, the method has taken: start ten a day, and put the reviews first.