How we measure reading levels

Every edition in the library carries a label such as B2.1. A program calculates it from the text itself: nobody types it in, and the level an adaptation was meant for does not count. The letter and the first digit are the CEFR band you know from courses and exams. The decimal shows where inside that band the text sits: B1.2 is near the easy end of B1, B1.8 near the hard end, and B1.5 is a typical B1 exam reading text.

What the label tells you

• The band comes first. B1.0 and B1.9 are both B1. The decimal is a finer position on the same scale, useful for comparing two editions.
• The label and the catalogue section are different things. The B1 section holds our simplest retelling of a book; the label measures its text. Our B1-section editions usually measure between B1.8 and B2.5: they follow the original chapter by chapter and are denser than the section name suggests.
• Originals are measured on the same scale. If a novel measures C2 in the original, its B2 edition might come out at C1.1 and its B1 edition at B2.1. We only publish an adaptation that is clearly easier than its original.
• C2 has no decimal. Neither our program nor the other checks we tried tell C1 texts from C2 texts reliably, so everything harder than C1.9 is simply labelled C2.
• Small differences mean nothing. B1.6 and B1.7 read the same in practice. A difference of half a band or more you will notice.

How the program measures a text

It splits the book into passages of about 220 words and looks at each one in three ways.

The passages are combined into one score for the whole book, and the score becomes the label: 4.31 is B2.3. Words explained in brackets are measured together with their explanation, the way you read them.

How we set the scale

The weights and the conversion from score to label were fitted on 200 reading texts from the free sample papers of four exam boards — Cambridge English, LanguageCert, Pearson and Trinity College London — from A2 to C2: a typical exam text of a band gets the middle of that band (B2.5 for B2). We tested it fairly: each time the program was fitted without one exam board and then measured that board's texts unseen.

Exam texts are articles and letters; our books are novels. So we checked separately whether the program confuses genres, on 4,724 passages whose difficulty teachers rated by comparing them in pairs^[4]. At the same teacher rating, the program placed literary passages slightly lower — by 0.13. We raised book labels by that amount.

The method has a version number, and a version in use never changes: the same text always gets the same label. When we improve the method, a new version comes out and every edition in the library is measured again. The current version is 1.2 (since 13 October 2026).

How we check the scale

A scale is only as good as its agreement with texts whose level is already known.

What this shows. Up to C1 our labels match the exam level within half a band. The program does not tell C2 texts from C1 texts, which is why C2 has no decimal. Graded readers from the Cambridge, Penguin and Helbling series at B2 and C1 measure upper B1 on our scale. That is not a disagreement but a different question: a series labels a book by the reader who will find it easy, while our label measures the text itself, the way an exam measures a reading text.

We also checked the program against two independent methods. Language models from other developers compared passages for difficulty and agreed with the teachers even better than the program does; they too placed our editions above series books of the same band. A vocabulary count shows the same: in our B1 editions 2.2% of the words fall outside the 2,000 most frequent, against 1.5% in series B1 books and 2.5% in series B2 books.

The originals of classic novels measure between C1.2 and C2 on our scale. The CEFR itself puts them there too: reading “classical” literary writing is described at C2, “contemporary” literary texts at C1, and novels in “straightforward, unelaborated language” at B2^[5].

The medians rest on hundreds of texts, so they are stable; a single short text can still differ from its level by up to a band, which is why we label whole books, not passages. A book's label is accurate to about half a band.

What the label does not tell you

• It measures the text, not you. To find your own level, take the vocabulary test and start with an edition of your band.
• It is not an exam result. No program can certify a text the way an exam board certifies a candidate; the label is a careful estimate on one consistent scale.
• It does not measure the story. Two books with the same label can feel different: an old novel keeps its period objects and customs. The words a story needs are kept and explained in brackets where they first appear.

How to choose a book at your level and actually finish it is in how to read books in English.

Questions