Most kanji have two families of readings. On readings came from Chinese when the characters were borrowed, and kun readings are native Japanese words attached to the character’s meaning. Knowing which applies in a given word is one of the genuine difficulties of Japanese, and it follows patterns rather than being arbitrary.
Where the two families come from
Kanji arrived in Japan from China along with their pronunciations. Those borrowed pronunciations, adapted to Japanese sounds, became the on readings, written in katakana in dictionaries.
Japanese already had its own words for the things the characters represented. Attaching an existing native word to a borrowed character produced the kun reading, written in hiragana in dictionaries.
So 水 has an on reading, sui, from Chinese, and a kun reading, mizu, which is the Japanese word for water that existed before the character arrived. Both are correct, and which one you use depends on the word.
Some kanji have several of each, because readings were borrowed at different periods from different regions of China, which is why 行 can be kou, gyou, an, iku, okonau and more depending on context. The jōyō kanji table published by the Agency for Cultural Affairs lists the readings considered standard for general use, which is a useful boundary when a dictionary offers more than you need.
| Reading type | Origin | Written in dictionaries as | Example for 水 |
|---|---|---|---|
| On reading | Borrowed from Chinese | Katakana | sui |
| Kun reading | Native Japanese word | Hiragana | mizu |
| Multiple on readings | Different borrowing periods | Katakana | Common in older kanji |
| Nanori | Name readings | Varies | Used in personal names |
The patterns that mostly hold
Three rules of thumb get you a long way, and each has exceptions.
A kanji standing alone as a word usually takes its kun reading. 水 alone is mizu. 山 alone is yama.
A kanji in a compound with other kanji usually takes its on reading. 水曜日 is suiyoubi, 火山 is kazan. This is the single most useful pattern, because compounds are everywhere in written Japanese.
A kanji followed by hiragana endings usually takes a kun reading, because those endings are inflections of a native Japanese word. 食べる is taberu, 読む is yomu. Those trailing kana are okurigana, and they are part of how the reading is signalled.
The exceptions are real and worth expecting rather than resenting: 毎日 is mainichi with an on reading, while 今日 is kyou, which is neither a straightforward on nor kun reading of its characters. Readings of that last kind, assigned to a whole word rather than derived character by character, are common enough in everyday vocabulary that they are learned as words from the start.
Why this matters for writing
You do not choose readings when you write by hand; you write what the word is. The reason readings matter for a writer is different and more practical: they determine what you can write from memory when prompted.
If you practise with a prompt like “water”, the ambiguity is immediate: 水 alone, or the water in 水曜日? A prompt using the actual word, “mizu, water” or “suiyoubi, Wednesday”, removes it and rehearses something you will use.
This is the argument for practising vocabulary rather than isolated characters once you are past the first hundred or so. Words carry their reading, their meaning and their context, and the kanji comes along inside them.
There is also a spelling dimension. Choosing the right kanji for a word you know by sound is a real skill in Japanese, since homophones are common, and it is trained by writing words rather than by writing characters.
Where readings bite hardest for learners
Three situations account for most of the frustration, and each has a practical response.
Names. Personal and place names use readings that appear nowhere else, including readings reserved for names, and there is no reliable way to derive them from the characters alone. Native speakers ask, routinely, and business cards exist partly for this reason. The correct response is to treat name readings as vocabulary rather than as a failure of your kanji knowledge.
Homophone choice. Japanese has many words that sound identical and are written with different kanji, so knowing a word by sound does not tell you how to write it. This is a writing problem specifically, and it is trained by writing words rather than characters, which is one more argument for practising vocabulary.
Compound readings that break the pattern. Some compounds take kun readings, some mix the two, and some are assigned wholesale to a word. These are learned individually, and the useful mindset is that they are a small fraction of a large regular system rather than evidence that the system has no rules.
There is also a fourth, quieter difficulty: the same kanji taking a different reading when it is part of a longer compound, sometimes with a sound change at the boundary. Those changes follow patterns of their own, and they are absorbed through exposure rather than memorised in advance. Trying to learn them as rules before meeting them in words is a common way to make Japanese feel harder than it is.
Learning readings without drowning
Do not learn all readings for a kanji at once. A dictionary entry may list six, most of which appear in words you will not meet for years. Learn the readings that come with the words you are learning, and let the rest arrive with their words.
Use the compound pattern actively. When you meet a new two-kanji compound, expect on readings, and check rather than guess. The habit builds an ear for which reading sounds right, which is most of what fluent readers rely on.
Learn phonetic components. Many kanji share a phonetic part that predicts the on reading: 青 sei in 清, 晴 and 静. This is the closest Japanese comes to a pronunciation rule for kanji, and it is worth noticing when it appears.
And accept a period of uncertainty. Everyone learning Japanese guesses readings wrong for a long time, including on common words, and it resolves through volume of reading rather than through memorising tables.
| Situation | Expect | Example |
|---|---|---|
| Kanji alone as a word | Kun reading | 山 yama |
| Two or more kanji together | On reading | 火山 kazan |
| Kanji plus hiragana ending | Kun reading | 読む yomu |
| Personal and place names | Anything, including nanori | 名前 readings vary widely |
| Recently coined compounds | On reading | Technical vocabulary |
What the JLPT expects of readings
For learners preparing for an exam, it is worth knowing exactly what is being tested, because it changes how much reading detail to memorise.
The JLPT level summaries describe expected competence in reading and listening rather than listing readings to memorise, and the test itself is multiple choice. That means readings are tested in the sense that you must understand words containing them, and not in the sense that you must recite every reading of a character.
The practical implication is reassuring. A learner who knows the words is fine; a learner who memorised reading tables without the vocabulary is not, since the questions are about words in context rather than about characters in isolation.
For handwriting there is no exam pressure at all, since the JLPT contains no writing section. That makes handwriting a goal you choose rather than one imposed, and it is worth choosing deliberately rather than assuming an exam will force it.
The practical version for a writer
Practise words rather than characters. Prompt with a word and its meaning, write the kanji from memory, and check the whole word including any okurigana.
Keep the reading with the item. A card or prompt that carries “yomu, to read” teaches you something usable; one that says only “read” teaches you an ambiguity. The same applies to okurigana: practising 読 alone and 読む as a word are different exercises, and only the second matches what you will actually write.
Write compounds. They are how kanji actually appear, they rehearse on readings, and they give each character a context that makes it stick. A compound also catches errors that isolated practice never surfaces, since two characters written side by side expose spacing and proportion problems that a single character in the middle of a page hides. Doing it by hand rather than by keyboard also appears to strengthen the memory: a brain-wave study found words learned by handwriting were encoded more strongly than the same words typed.
{{appName}} works this way: prompt first with the word, character hidden, stroke order checked on your attempt, misses returned sooner. It is free in early access. For the surrounding method, practising writing kanji and kanji compounds practice go further.
What to do when a reading will not stick
A reading that keeps escaping is usually failing for a nameable reason.
If you know the character and not the reading, the character was learned in isolation. Attach it to a word you say and the reading arrives with the word, which is why vocabulary-first study produces fewer of these gaps.
If you keep producing the on reading where a kun is needed, or the reverse, the pattern rules have not become automatic yet. Say the pattern out loud when checking: “compound, so on reading”, “kana ending, so kun reading”. A few dozen deliberate applications makes it habitual.
If a word has an assigned reading that ignores its characters, stop analysing it and learn it as a word. Trying to derive 今日 from its characters is wasted effort, and treating it as vocabulary takes seconds.
And if several readings are competing, check whether you actually need all of them yet. Most learners are carrying readings they met in a dictionary rather than in a sentence, and dropping the ones without a word attached lightens the load considerably.
Frequently asked questions
What is the difference between on and kun readings?
On readings are pronunciations borrowed from Chinese when the kanji were imported, written in katakana in dictionaries. Kun readings are native Japanese words attached to the character’s meaning, written in hiragana. 水 has the on reading sui and the kun reading mizu, and which one applies depends on the word the character appears in.
How do I know which reading to use?
Three patterns cover most cases. A kanji standing alone as a word usually takes its kun reading. A kanji in a compound with other kanji usually takes its on reading. A kanji followed by hiragana endings usually takes a kun reading, because those endings inflect a native Japanese word. Exceptions exist and are learned with their words.
Why do some kanji have so many readings?
Because readings were borrowed from different regions of China at different periods, and native Japanese words were attached separately. A character like 行 accumulated several on readings from successive borrowings plus multiple kun readings. Names add another layer, since personal and place names use readings that appear nowhere else.
Do I need to memorise every reading of a kanji?
No, and trying to is a common way to stall. Learn the readings that come with the words you are actually learning, and let the rest arrive with their words. A dictionary entry listing six readings is describing the character’s whole history, not a set you need this month.
What is the best app for practising kanji with their readings?
{{appName}} prompts with the word and its reading rather than an isolated character, keeps the kanji hidden until you have written it, checks the stroke order of your attempt and brings back what you miss. Practising words rather than characters is what keeps readings attached to something usable. It is free in early access.
