Chinese words in this text can be activated to show a dictionary gloss.
Most people who say they cannot remember Chinese characters are not describing a memory problem. They are describing what happens when you try to memorise something you have never analysed. A character met as a bundle of strokes is close to unmemorable: there is nothing for the memory to grip, and the twelfth one you learn looks like the eleventh. The same character met as an arrangement of two or three components you already know — one of which usually tells you roughly how it sounds — is ordinary vocabulary, no harder than a long word in any other language.
So the work is not memorising harder. It is learning to see what is actually in front of you.
Characters are made of parts, and the parts repeat
Only a few hundred components are in regular use, and they recombine endlessly. , , , , , all begin with , the compressed form of (speech). , , , , all begin with , the compressed form of (water). Once you see the component rather than the strokes, a character that looked like scribble becomes two or three chunks — and short-term memory holds chunks, not strokes.
A useful habit is to stop asking "how many strokes is this?" and start asking "what is this made of?". (xiè) is not thirteen strokes; it is plus (shè). (xiǎng, to think, to want) is not thirteen strokes; it is (xiāng) on top of (heart) — and the top half is telling you the sound while the bottom half is telling you the domain.
The most useful components to know early are the ones that appear constantly: person, speech, water, heart, hand, mouth, woman, tree, eye, metal, food, silk, illness, roof, grass, movement.
Most characters are a meaning part plus a sound part
This is the single fact that changes the work most, and it is systematically underused by learners. Most Chinese characters are not pictures. They are compounds in which one part indicates the semantic field and the other indicates, approximately, the pronunciation.
Learn to read them in series. (qīng) is the sound part in a whole family:
- qǐng (to invite, please) — speech
- qīng (clear) — water
- qíng (feeling) — heart
- qíng (fine weather) — sun
- jīng (eye, as in yǎnjing) — eye
The same pattern with (bāo): bào (to hug, hand), pǎo (to run, foot), pào (to steep, water), bǎo (full, food), pào (cannon, fire). And with (mǎ): mā (mother, ), ma (question particle, ), mà (to scold, two ), mǎ (number, as in hàomǎ, stone).
Notice what the series shows: the sound part gives you the syllable, not the tone. That is normal, and it is the first honest limit to state.
The second limit is bigger. The sound parts were fixed thousands of years ago and the language has moved since, so a fair number of series are now loose. (gōng) sits inside (jiāng, river), (hóng, red), (kōng, empty) and (gōng, merit). You would never derive jiāng from gōng. Treat the sound part as a strong hint that narrows the field and occasionally as a precise answer — never as a rule.
Even so, the question "have I seen this sound part before?" is worth asking of every new character, because when it pays it pays enormously: one insight about disposes of five characters.
Mnemonics that work
A mnemonic is a temporary bridge between a shape and a meaning. Good ones share four properties.
They are built from the real components, using the real meanings. (xiū, to rest) is person beside tree: a person resting against a tree. (míng, bright) is sun and moon. (hǎo, good) is woman and child. (sēn, forest) is three . (jiā, home) is roof over pig. These stories work because they survive contact with the next character that shares a component.
They include the sound. A mnemonic that gets you to "illness" for and leaves you unable to say bìng has done half a job. Better: is the illness component and is bǐng — the meaning is on the left and the sound is on the right, one tone away.
They are specific and slightly absurd. Vague images decay. A precise, odd, sensory image does not.
They are yours. A mnemonic you generated will outperform one you read in a book, because generating it is itself a retrieval event.
Mnemonics that collapse
They fail in predictable ways, and the failures are worth naming so you can recognise them early.
Stories built from stroke shapes rather than components. "A little man wearing a hat carrying a stick" describes hundreds of characters equally well. These stories collide with each other within weeks, and when two mnemonics collide you lose both.
Stories with no sound attached. You end up with a large silent vocabulary: characters you can read on paper and cannot say, hear or use.
Stories that fight the real structure. If your invented tale for makes no room for and , it will be contradicted the moment you meet , and , and you will have to unlearn it.
Stories that are too long. If retrieving the mnemonic takes longer than retrieving the character would have, the scaffolding has become the building.
One more thing worth saying plainly: a great many popular character stories are folk etymology rather than history. That is not a disaster — a bridge does not have to be true to hold your weight — but you should know which of your stories are inventions, never repeat them to anyone as fact, and expect to drop them. A mnemonic is scaffolding, and scaffolding is supposed to come down. Once arrives instantly, the story about it is dead weight.
Where rote copying genuinely belongs
Copying characters by hand has a bad reputation among efficiency-minded learners, and the reputation is only half deserved. Copying does three jobs that nothing else does as well.
It teaches stroke order ( bǐshùn), which matters more than it looks: correct order produces correct proportions, makes handwriting legible, and is what handwriting-recognition input expects. The basic rules cover most cases — top before bottom, left before right, horizontal before crossing vertical (), outside before inside, and close the box last ().
It builds motor memory, which is the only route to writing from memory rather than recognising from a list.
And it forces discrimination between forms that look identical until your hand has made both: / / , / , / , / , / , / / , / , / , / .
What copying does not do is build recall, and this is where the traditional method wastes time. Writing fifty times in one sitting trains your hand for the first five repetitions and then trains your attention to wander. Five careful copies today, five tomorrow and five on Friday will beat fifty in a row, for exactly the reason spaced review beats cramming.
Two refinements make copying far more efficient. Copy words, not characters: write , , rather than isolated forms, so the character arrives attached to something you will actually say. And copy from memory, not from the model: look at the character, cover it, write it, then check. Copying while looking is tracing, and tracing teaches almost nothing.
A loop for each new character
Meet it inside a word, inside a sentence — never as a bare form on a list. Break it into components and name them. Ask whether the sound part belongs to a series you know. Invent one short story only if the structure does not already explain itself. Write it from memory three to five times, correcting against the model. Make a card that asks for sound and meaning together, and count the card as failed if the reading was wrong. Then go and read something that contains it.
That last step is the one that decides everything. Characters are not retained by the effort that introduced them; they are retained by being met again, in text, where the sentence is doing some of the remembering for you.
What to expect
The first few hundred characters are the hardest, because you are learning the components and the characters at the same time. After that the curve bends, and it bends noticeably: new characters increasingly turn out to be recombinations of things you already know, and your rate of learning rises even as the characters get rarer.
The sign that it is working is not a number. It is the moment you meet an unfamiliar character, guess its approximate sound from the right-hand side and its general domain from the left, and find you are roughly right.
About this article
- Author
- Mehdi Chraïbi
- Version
- 1
- Published
- January 12, 2026
- Last verified
- Not recorded on this article
Articles are versioned. If a correction is needed the piece is revised and the version increases — it is not quietly rewritten or deleted. Report an error
Independent HSK preparation platform. Not affiliated with or endorsed by Chinese Testing International or the official HSK examination authorities.