On'yomi vs kun'yomi: how to know which kanji reading to use

Every learner hits the same wall around their fiftieth kanji: the character they memorised as やま turns up in a word and is suddenly さん. Nothing in the shape of 山 announces which reading applies, and the dictionary lists both without telling you when to use which. It looks arbitrary. It mostly isn't. Two simple rules predict the reading in the large majority of everyday words, a handful of sound changes explain most of the leftovers, and the genuinely irregular cases form a short list you can memorise in an afternoon. This guide lays out that system with worked examples, and finishes with the practical question nobody answers: what do you actually do when you meet a kanji you can't read?

1. Where the two readings come from

Japan had a spoken language long before it had a writing system. When Chinese characters arrived — over several centuries, from different regions of China, in several waves — they came as a package: a written symbol, a meaning, and a Chinese pronunciation. Japanese absorbed all three, but in two different ways.

The on'yomi (音読み) is the imported reading: a Japanese approximation of how the character sounded in Chinese. The kun'yomi (訓読み) is the native reading: the Japanese word that already existed for that meaning, attached to the borrowed symbol. So 山 arrived carrying a Chinese sound that became サン, and it was also glued onto the existing Japanese word やま, which had been in the language all along. Both readings are equally "correct"; they simply belong to different layers of the vocabulary.

This is the same split English speakers live with, largely without noticing. "Water" is native English, but the medical adjective is "aqueous"; "eye" is native, but the specialist is an "ocular" surgeon. The everyday word is Germanic, the technical word is Latinate. Japanese does the same thing with kun'yomi and on'yomi, and the register difference carries over: on'yomi compounds sound more formal, technical and abstract, while kun'yomi words feel concrete and everyday. 水 (みず) is the water you drink; 水道 (すいどう) is the water supply system.

One typographic convention will help you read dictionary entries: on'yomi are conventionally written in katakana and kun'yomi in hiragana. When a reference shows 山 as「サン / やま」, it is telling you the first is the Chinese-derived reading and the second is native, without spending a word on it. The same convention appears throughout Japanese dictionaries and reference tables.

Because the borrowings happened in waves, a single character can carry more than one on'yomi from different eras. 明 has メイ and ミョウ; 人 has ジン and ニン; 月 has ゲツ and ガツ; 生 has セイ and ショウ. You do not need to learn which historical wave produced which — you need to learn the words. That point comes back in section 8, and it is the single most useful idea in this guide.

2. Rule one: two or more kanji together usually means on'yomi

Compound words made of kanji stuck directly together — no kana between them — normally take on'yomi throughout. This is the workhorse rule, and it covers a very large share of the written vocabulary, because Sino-Japanese compounds are how Japanese builds technical, academic and news language.

• 山 alone is やま, but 火山 is かざん (volcano), 登山 is とざん (mountain climbing), 富士山 is ふじさん.
• 水 alone is みず, but 水曜日 is すいようび (Wednesday), 水泳 is すいえい (swimming), 香水 is こうすい (perfume).
• 学 rarely stands alone, and its compounds are all on: 学校 がっこう (school), 大学 だいがく (university), 学生 がくせい (student), 科学 かがく (science).
• 電 + 車 → 電車 でんしゃ (train); 電 + 話 → 電話 でんわ (telephone); 自 + 転 + 車 → 自転車 じてんしゃ (bicycle). Three kanji, three on'yomi.

The rule has a well-behaved exception: native compounds, where two kanji represent a word that was always Japanese. These take kun'yomi on both halves. 手 (て) + 紙 (かみ) gives 手紙 てがみ, "letter". 花 (はな) + 見 (み) gives 花見 はなみ, "cherry-blossom viewing". 山 (やま) + 道 (みち) gives 山道 やまみち, "mountain path". 昼 (ひる) + 休み (やすみ) gives 昼休み ひるやすみ, "lunch break". You can often spot these because the meaning is everyday and physical rather than abstract.

And then there are the genuine mixtures, which Japanese has names for. 重箱読み (じゅうばこよみ) is on + kun: 重箱 じゅうばこ itself (a stacked lacquer box), 台所 だいどころ (kitchen), 番組 ばんぐみ (TV programme). 湯桶読み (ゆとうよみ) is kun + on: 湯桶 ゆとう, 場所 ばしょ (place), 手本 てほん (model, example), 見本 みほん (sample). These are a minority, but they are common words, so treat them as vocabulary rather than as a rule failure.

3. Rule two: a kana tail means kun'yomi

When a kanji is followed by hiragana that inflects — the okurigana (送り仮名) — you are looking at a native Japanese word, and the kanji takes its kun'yomi. Verbs and adjectives almost always work this way, which is why they behave so differently from compound nouns.

• 食べる → たべる (to eat), but 食事 → しょくじ (a meal) and 食堂 → しょくどう (dining hall).
• 行く → いく (to go), but 旅行 → りょこう (travel) and 銀行 → ぎんこう (bank).
• 新しい → あたらしい (new), but 新聞 → しんぶん (newspaper).
• 楽しい → たのしい (fun), but 音楽 → おんがく (music).

Okurigana does more than signal the reading — it distinguishes different words written with the same character. 上 gives you 上がる (あがる, to go up, intransitive) and 上る (のぼる, to climb). 下 gives 下がる (さがる, to drop), 下りる (おりる, to descend, to get off), and 下さい (ください, please). 生 gives 生きる (いきる, to live), 生まれる (うまれる, to be born) and 生える (はえる, to grow, of hair or plants). The kana tell you which word you're reading, so learners who skip okurigana when writing are throwing away the only disambiguation the script provides.

A third, weaker signal is worth knowing: a single kanji standing alone as a whole word is usually kun'yomi — 山 やま, 川 かわ, 水 みず, 人 ひと, 花 はな, 手 て. But plenty of single characters are on'yomi words in their own right, especially abstract or borrowed concepts: 本 ほん (book) is on, 駅 えき (station) is on, 絵 え (picture) is on, 茶 ちゃ (tea) is on. So treat "alone = kun" as a first guess, not a law.

4. The same kanji in both modes — worked examples

Rules are easier to trust once you've watched them work. Here are seven very common kanji, each shown in kun'yomi words and in on'yomi compounds. Read across each row and you'll see the two rules doing all the work.

Look at 月 for a moment, because it shows something the rules alone don't: the same character takes ゲツ in 月曜日 and 今月 but ガツ in 一月 and 正月. Both are on'yomi, from different borrowing layers, and there is no way to derive which from the shape. You learn 一月 as a word.

生 is the famous extreme case, and it is worth confronting rather than avoiding. Between its on'yomi (セイ・ショウ) and its many kun'yomi (い-, う-, は-, なま, き), this one character participates in dozens of distinct words: 学生 がくせい, 先生 せんせい, 生活 せいかつ (daily life), 人生 じんせい (human life), 一生 いっしょう, 誕生日 たんじょうび, 生きる いきる, 生まれる うまれる, 生える はえる, 生ビール なまビール, and the outright irregular 芝生 しばふ (lawn). Nobody memorises that as a list of readings. Native speakers know the words, and the readings come along for free. That is the model to copy.

5. Sound changes: rendaku and the small っ

A lot of apparent irregularity is actually regular phonetic adjustment. Two processes account for most of it.

Rendaku (連濁), sequential voicing. When two elements join into one word, the first consonant of the second element often becomes voiced: か→が, さ→ざ, た→だ, は→ば or ぱ. 花 (はな) + 火 (ひ) becomes 花火 はなび, "fireworks". More examples with the source words visible:

• 手 (て) + 紙 (かみ) → 手紙 てがみ (letter)
• 時 (とき) + 時 → 時々 ときどき (sometimes); 人 (ひと) + 人 → 人々 ひとびと (people)
• 青 (あお) + 空 (そら) → 青空 あおぞら (blue sky)
• 昔 (むかし) + 話 (はなし) → 昔話 むかしばなし (old tale)
• 折り (おり) + 紙 (かみ) → 折り紙 おりがみ; 小 (こ) + 包 (つつみ) → 小包 こづつみ (parcel)
• It happens to on'yomi too: 会社 (かいしゃ) inside 株式会社 becomes かぶしきがいしゃ (joint-stock company).

Rendaku is a tendency, not a guarantee, and it has at least one solid brake on it, known as Lyman's Law: if the second element already contains a voiced obstruent, rendaku is normally blocked. That is why 春 (はる) + 風 (かぜ) stays 春風 はるかぜ rather than becoming ×はるがぜ — かぜ already has a ぜ. Beyond that, whether a given compound voices is largely lexical, so the honest advice is: expect it, recognise it when you hear it, and learn the actual word.

Gemination, the small っ. When an on'yomi ending in く, ち or つ is followed by another on'yomi starting with an unvoiced consonant, the ending collapses into a small っ and the following sound often hardens. 学 (がく) + 校 (こう) → 学校 がっこう, not ×がくこう. The same process gives:

• 日 (にち) + 記 (き) → 日記 にっき (diary)
• 雑 (ざつ) + 誌 (し) → 雑誌 ざっし (magazine)
• 出 (しゅつ) + 発 (はつ) → 出発 しゅっぱつ (departure) — note the は hardening to ぱ
• 発 (はつ) + 表 (ひょう) → 発表 はっぴょう (presentation, announcement)
• 決 (けつ) + 定 (てい) → 決定 けってい (decision)

This one matters for listening as much as reading: the pause of a small っ is a real, timed beat in Japanese, and getting it wrong changes the word. If you haven't drilled that yet, the pronunciation guide covers how to hold it properly.

6. Jukujikun: whole words with unpredictable readings

Some words are written with kanji chosen for meaning, with a reading that belongs to the word as a unit and cannot be split across the characters. These are 熟字訓 (じゅくじくん). You cannot derive 今日 = きょう by assigning sounds to 今 and 日 separately — the whole compound simply reads きょう. Here are the ones you will meet earliest and most often:

Notice how many of these are absolute beginner vocabulary — 今日, 明日, 昨日, 一人, 二人, 上手, 下手 all appear in the first weeks of study and in the N5 word list. That is not bad luck; the most irregular words in any language tend to be the most frequent ones, because frequency protects them from being regularised. The upside is that they arrive early, you use them constantly, and they stop feeling irregular fast.

A related category worth flagging: names. Personal names and place names use readings (名乗り, なのり) that are not always available to ordinary vocabulary, and the same written surname can be read two different ways by two different families. Japanese forms routinely include a furigana field for exactly this reason. Do not treat a name you can't read as a failure of your kanji study — often even a native speaker has to ask.

7. Numbers and counters: a special zone

Numbers deserve their own warning label, because they break the tidy on/kun split in a way nothing else does. Several digits have two or three competing readings, and which one appears depends on the counter that follows — sometimes for reasons of euphony that go back centuries.

• 四 is し or よん or よ: 四月 しがつ (April), 四人 よにん (four people), 四時 よじ (four o'clock).
• 七 is しち or なな: 七月 しちがつ (July), 七時 しちじ (seven o'clock), but なな is the everyday counting form.
• 九 is きゅう or く: 九月 くがつ (September), 九時 くじ (nine o'clock), 九人 きゅうにん (nine people).
• Dates are a self-contained system with native readings: 一日 ついたち (1st), 二日 ふつか, 三日 みっか, 四日 よっか, 五日 いつか, 六日 むいか, 七日 なのか, 八日 ようか, 九日 ここのか, 十日 とおか, 二十日 はつか (20th). Note that 一日 means ついたち as a date but いちにち as "one day" — the same two characters, two readings, two meanings.
• Counters trigger the sound changes from section 5: 一本 いっぽん, 三本 さんぼん, 六本 ろっぽん; 一匹 いっぴき, 三匹 さんびき, 六匹 ろっぴき; 一階 いっかい, 三階 さんがい, 六階 ろっかい; 一分 いっぷん, 三分 さんぷん.

There is no shortcut here, and trying to derive these from rules wastes time. Counters are learned as sound patterns — say them out loud in sequence until the rhythm is automatic. The counters guide works through the main ones properly, including which counter goes with which kind of object.

8. Learn readings inside words, never as lists

Here is the practical conclusion the whole guide has been building toward. A kanji flashcard that says「生 — セイ、ショウ、い(きる)、う(まれる)、は(える)、なま、き」is technically accurate and nearly useless. You cannot produce a sentence with it, you cannot recognise it in text with it, and it takes as long to review as five real words would.

Learn 学生 (がくせい, student), 先生 (せんせい, teacher), 生まれる (うまれる, to be born) and 誕生日 (たんじょうび, birthday) as four separate vocabulary items instead. You get four usable words, and the readings of 生 arrive as a by-product, correctly bound to the contexts where they apply. After twenty or thirty such words you will start guessing new compounds correctly — 生活, 人生, 一生 — because you have internalised a distribution rather than a list. That is exactly how native speakers acquire it.

Two things make this approach work better in practice:

1. Cluster words that share a kanji. When you meet 校 in 学校, look at 高校 (こうこう, high school) and 校長 (こうちょう, principal) in the same sitting. The shared コウ reading gets three anchors instead of one, at almost no extra cost. Jotting down the words that share a character as you meet them is enough to build the habit.
2. Space the reviews. Readings are exactly the kind of arbitrary sound-symbol link that decays fast and recovers cheaply — the classic use case for spaced repetition. Feed the words into your review deck; the spaced repetition guide explains how to set the intervals so you're not re-learning the same reading every week.

For scale: the 常用漢字 (じょうようかんじ) list — the characters designated for everyday use — contains 2,136 characters, and most of them have two or more readings. Approached as a reading table, that is a punishing amount of memorisation. Approached as vocabulary, it is simply what happens when you learn a few thousand words, which you were going to do anyway. Working level by level through the JLPT module or the basic vocabulary module builds the same coverage without ever making readings the unit of study.

9. What to do with a kanji you can't read

This will keep happening for years, so it is worth having a routine instead of a feeling of defeat.

1. Guess from the phonetic component. Around half of all kanji are phono-semantic compounds: one part hints at meaning, the other at sound. Once you know 青 is セイ, you have a decent shot at 晴 (せい, as in 晴天 せいてん), 清 (せい, 清潔 せいけつ) and 精 (せい, 精神 せいしん). Once you know 交 is コウ, try 校 (こう) and 効 (こう). The trick is real but leaky — 寺 is ジ and gives you 時 じ and 持 じ, yet 待 is タイ. Use it as a first guess, then verify.
2. Look it up by radical. If the character is in print and you can't type it, identify the radical and the stroke count and search that way. This is the oldest method and still the most reliable for handwriting; the radicals guide covers the components worth knowing by name.
3. Draw it. Every phone IME has handwriting input, and the recognisers are forgiving about stroke order. Sketching an unknown character is usually faster than radical lookup, and you can paste the result straight into the dictionary.
4. Read the context before the character. If it's a two-kanji compound and you know one half, you can often narrow the meaning enough to keep reading. Fluent reading means tolerating gaps; stopping at every unknown character is how people abandon their first novel.
5. Record it as a word, not a character. When you do look it up, save the whole word with its reading and an example, not the isolated kanji. This is the same principle as section 8, applied at the moment it matters most.

One last reassurance about how far the rules take you. In a typical newspaper sentence, the compound-noun rule and the okurigana rule between them predict the reading of most characters correctly. The residue — jukujikun, counters, names, the occasional unexplainable compound — is real, but it is a finite list rather than an infinite fog, and the highest-frequency members of it are all in section 6 above. If you can read a restaurant menu, you already handle a chunk of this without thinking — menus are a good low-stakes place to watch these rules operate on real text, because the same fifty or so food kanji recur endlessly in both reading modes.