研究 · from our answer data

The hardest JLPT question types: what 11,779 answers show

Which JLPT question formats do learners actually get wrong? We looked at 11,779 first answers from 527 learners practising N5 to N1 on this site, across 16 of the exam’s question formats.

The short answer

The hardest format is Context fill (文脈規定): choosing the word that fits a sentence. Learners got 55.5% right on their first try. Next come Text grammar (58.1%) and Usage (63.3%). The easiest are the reading passages, at 78.3–83%.

The pattern underneath: what trips learners up is not meaning but fit, which words go together and how sentences join. Knowing what a word means is not the same as knowing where it goes.

All 16 question types, hardest first

Share of first answers that were right. The shaded band is the 95% range the true figure most likely sits in; types with fewer answers have wider bands.

  1. 1Context fill 文脈規定55.5%1,193 answers · 259 learners · range 52.7–58.3%
  2. 2Text grammar 文章の文法58.1%677 answers · 166 learners · range 54.3–61.7%
  3. 3Usage 用法63.3%539 answers · 166 learners · range 59.1–67.2%
  4. 4Grammar form 文法形式の判断64.8%2,309 answers · 294 learners · range 62.8–66.7%
  5. 5Sentence build ★ 文の組み立て65%814 answers · 205 learners · range 61.6–68.2%
  6. 6Utterance expression 発話表現66.7%282 answers · 81 learners · range 61–71.9%
  7. 7Paraphrase 言い換え類義68.5%731 answers · 231 learners · range 65.1–71.8%
  8. 8Task comprehension 課題理解68.9%492 answers · 115 learners · range 64.7–72.8%
  9. 9Quick response 即時応答70.7%351 answers · 107 learners · range 65.7–75.2%
  10. 10Orthography 表記72.9%949 answers · 226 learners · range 70–75.6%
  11. 11Point comprehension ポイント理解74.3%401 answers · 113 learners · range 69.8–78.3%
  12. 12Info retrieval 情報検索78.3%318 answers · 145 learners · range 73.4–82.5%
  13. 13Mid passage 内容理解(中文)80%450 answers · 133 learners · range 76.1–83.4%
  14. 14Kanji reading 漢字読み80.8%1,444 answers · 322 learners · range 78.7–82.8%
  15. 15Long passage 内容理解(長文)81.9%160 answers · 68 learners · range 75.2–87.1%
  16. 16Short passage 内容理解(短文)83%669 answers · 192 learners · range 79.9–85.6%

1. Vocabulary in context is the hardest thing on the paper

Context fill and Kanji reading are both in the vocabulary section, at the same levels. Yet learners read kanji words correctly 80.8% of the time and chose the right word for a sentence only 55.5% of the time, a gap of 25 points.

The most-missed questions show why. They are not rare words. They are words that have to go with a particular partner:

  • きってを はって、てがみを 【だします】。I put a stamp on the letter and post it. Only 15% of 20 learners chose だします. In Japanese you “put out” a letter (手紙を出す), and あけます (open) was the most common wrong pick. N5.
  • 新しい制度が来月から【じっし】される。The new system will be put into effect starting next month. 33.3% of 21 learners chose じっし (実施). The usual verb for a system coming into force is 実施; the options beside it, 実験 (experiment), 実感 (feeling) and 実現 (realising a goal), are all real words that sound just as plausible. N2.

Usage (63.3%) fails the same way from the other direction: it gives you the word and asks which sentence uses it naturally. With 思い切って (“taking the plunge”), 54.2% of 24 learners found the right sentence; the rest picked sentences where the word fits the grammar but not the meaning, such as closing a door “boldly and quietly”.

What to do: learn words in the phrases they come in (手紙を出す, 制度を実施する, 熱が下がる), not as single translations. Practise Context fill · how the format works

2. Grammar gets harder when it spans sentences

Text grammar (文章の文法) asks for grammar inside a passage rather than one sentence, and learners got 58.1% right, against 64.8% for single-sentence grammar form questions. The blank depends on what came before: a connector, a point of view, an intention. In one N5 text, a learner plans to buy a present for a party tomorrow: プレゼントを【かおう】とおもっています. Only 23.5% of 17 learners chose the intention form かおう; the plain かう was the usual pick.

What to do: read the sentences either side of every blank before looking at the options. Practise Text grammar · how the format works

3. Reading passages are the easiest, in practice

All four reading formats are among the five easiest, with kanji reading the only other format in that group: short passage 83%, long passage 81.9%, mid passage 80%, info retrieval 78.3%. That surprises people, because reading is the section many learners fear. Two cautions: practice here is untimed, and on exam day the reading section is hard mostly because of the clock. A learner who reads accurately but slowly will score lower there than these figures suggest.

The hardest type at each level

LevelHardestNext hardestAll types
N5Context fill 57.4%Utterance expression 60%71.7%
N4Text grammar 55.2%Context fill 57.5%70%
N3Text grammar 46.5%Usage 49.5%65%
N2Context fill 48.1%Grammar form 51.6%67%
N1Context fill 34.1%Paraphrase 40.5%61.5%

Two things stand out. N3 is a cliff for grammar: grammar form falls from 67.7% at N4 to 53.4% at N3, where grammar stops being particles and conjugations and becomes expressions such as とたんに and っぱなし. Across all types, learners scored lower at N3 (65%) than at N2 (67%), though people practising N2 here are likely stronger learners to begin with. And N1 vocabulary is the single hardest cell in the data: context fill at N1 was answered correctly 34.1% of the time, over 85 answers.

For beginners, listening is part of the problem: at N5, utterance expressions (60%) and task listening (60.4%) are the second and third hardest formats. Task listening then climbs to 79.4% at N4 and 87.2% at N3.

Most learners’ weakest type

Averages hide individuals, so we also looked at each learner on their own. Of the 76 learners with at least 5 answers in each of 4 or more types, 22 (29%) were weakest at Context fill. Next were Grammar form (13), Text grammar (10), Task comprehension (7). In all, 12 of the 16 formats were somebody’s weakest, which is why your own breakdown matters more than anyone’s average.

Fooled by a trap, or not knowing?

If learners were being tricked by one tempting wrong option, their wrong answers would pile up on it. They mostly do not. Across 281 questions with five or more misses, the most-picked wrong option drew 53.9% of wrong answers, against 51.7% if the same misses had been spread at random. Most wrong answers look like guesses: the learner did not know, rather than knew and was fooled. The clearest exception is orthography (56.5% against 50.2%), where a lookalike kanji really does pull people in.

That is good news: the fix is more words and more grammar, not exam tricks.

Practice moves the number

For the 25 cases where a learner answered 30 or more questions of one type, their accuracy went from 69.6% on their first five to 76.4% on the next ten and 79.5% on questions 16 to 30, always on questions they had not seen before. It is a small group of committed learners, so treat it as an encouraging sign rather than a promise.

Find your own weakest type

Every answer you give here is counted by question type. Your Me page ranks your types by accuracy at each level, next to how all learners do, and starts a practice set of just the one you pick. No account needed.

How this was measured

  • 21,789 answered questions were logged on JLPT Open Catalogue between 16 June 2026 and 19 November 2026. Only each learner’s first answer to each question counts, 11,779 in all, because a retry after reading the explanation says little about difficulty.
  • A learner is a signed-in account or, for people who are not signed in, an anonymous browser id. Answers with neither were left out. 527 learners in all; no personal details were used.
  • Ranges are 95% Wilson intervals. Level-by-type figures need at least 30 answers. The trap comparison simulates the same number of wrong answers spread at random over the same options.
  • The questions are this site’s own, written in the official formats, not past papers, and the learners chose what to practise. These figures describe learners on this site, not everyone who sits the JLPT, and they are not official JLPT statistics.
よくある質問 · Common questions

What is the hardest JLPT question type?

On JLPT Open Catalogue it is Context fill (文脈規定, contextually-defined expressions): learners answer 55.5% correctly on their first try, across 1,193 answers from 259 learners. Next are Text grammar (文章の文法, 58.1%) and Usage (用法, 63.3%). These are figures from learners on this site, not official JLPT data.

Is the JLPT reading section the hardest part?

Not in this data. The four reading formats are the easiest here, from 78.3% to 83% answered correctly, against 55.5% for context fill. Practice here is untimed, though, and the real reading section is hard largely because of time pressure.

Which JLPT question type should I practise most?

Your own weakest one, which varies by learner. Across 76 learners with enough answers to tell, the most common weakest type was Context fill (22 learners), followed by Grammar form (13) and Text grammar (10). The Me page on JLPT Open Catalogue ranks your own question types by accuracy.

己Me練Practice庫Library帳Decks録Progress