What ETS Actually Says
ETS publishes a good deal about how the TOEFL is built and scored: a technical manual, a specifications document for the 2026 update, an overview with worked examples of every task, and a guide to the new score scale.
This page reports what those four documents say, in plain terms, with each point attributed to the one it came from.
It also records where they are silent. Several things a student would reasonably want to know are simply not specified anywhere, and saying so is more useful than filling the gap with a confident number. Where we work to an estimate, we say it is an estimate.
Scores run 1 to 6, in half bands, and map onto CEFR.
A section score of 5 means C1 on that section — and it means the same thing on every section, which was not true before the January 2026 update.
The total score is an average, not a sum.
The four section scores are averaged and rounded to the nearest half band. A strong section cannot be added to the total; it can only pull the average up.
Listen and Repeat has 7 items, scored out of 5 each.
All seven are AI-scored, none machine-scored, with a CEFR target range spanning A1 to C2 — the widest of any task.
Listen and Repeat sentences get longer across a set.
The specification describes items "increasing in length and syntactic complexity", beginning with a single independent clause and building to compound verbs, dependent and relative clauses. The progression is required, not stylistic.
The scenario is given in writing as well as in audio.
The introduction that sets up a Speaking task is delivered both aurally and in print. Only the sentences to be repeated are audio-only.
Speaking tasks carry on-screen visuals.
The specification names "schematic images, maps, forms, or photos of speakers" as elements included to support understanding.
Announcements run about 40–85 words.
An explicit figure, and an easy one to miss: it appears in the Test Overview, not in the Technical Manual, which is silent on it.
Each section has a published time limit, and they are shorter than most people expect.
Reading 30 minutes, Listening 29, Writing 23, Speaking 8 — 90 minutes of base time in total. ETS adds that this excludes directions and that time varies as the test adapts, and asks test takers to allow about two hours overall.
The item counts ETS publishes are larger than the number scored.
The Test Content page lists 50 Reading and 47 Listening items; the Technical Manual scores 35 in each. The difference is unscored tryout questions, mixed in indistinguishably for quality control. Both figures are correct — they count different things.
Academic talks run about 175–250 words.
The Technical Manual gives 175–250. The Test Overview gives a looser "about 100–250". Where two ETS documents differ, the tighter figure is the safer target.
How long you get to speak in Listen and Repeat.
No response timing appears in any of the sources below — no number of seconds, no cue, no preparation period. Test takers describe a short window that closes by itself, and that account is consistent enough to plan around. It is simply not written down, so we practice against the clock rather than against a published figure.
How a repeated sentence is broken down when scored.
The documents state that all seven items are AI-scored and worth up to 5 points each. They do not publish the dimensions behind that score. Any named three-part breakdown you encounter comes from outside ETS.
How long a conversation runs.
Announcements and academic talks have stated lengths. Conversations do not. Two worked examples in the Test Overview measure 52 and 85 words, which gives a working range — but it is a measurement taken from examples, not a specification, and we treat it as such.
This is the question the scale change created, and the answer is arithmetic rather than ambition.
A single band on this scale covers what used to be a span of several points. An institution converting an existing requirement therefore cannot map it onto one band cleanly: choosing 5 would admit some applicants who fall below where its bar used to sit. Choosing 5.5 removes that risk.
The cost is that 5.5 is stricter than the old requirement really was, for applicants near the top of the 5 band. ETS names this trade-off in its own guidance — the higher bar reduces false admissions and increases false rejections.
So 5.5 is not a demand for exceptional C1. It is a cautious conversion onto a coarser scale, which is worth knowing if your target has moved up a band and the jump feels arbitrary.
| CEFR | Score |
|---|---|
| C2 | 6 |
| C1 | 5 – 5.5 |
| B2 | 4 – 4.5 |
| B1 | 3 – 3.5 |
| A2 | 2 – 2.5 |
| A1 | 1 – 1.5 |
- TOEFL iBT Test Content — the only ETS source that publishes section time limits. The Technical Manual, the Blueprint and the Test Overview are all silent on them. Checked 28 September 2026.
- TOEFL iBT Test: 2026 Update Test Blueprint and Specifications — the most specific source for how each task is built, and it states things the other documents do not.
- TOEFL iBT Technical Manual — Research Report RR-25-12, October 2025. The reference for test design and scoring.
- Updated TOEFL iBT Test Overview — worked examples of every task type, and the source of several lengths the Technical Manual omits. Large file; not one to open on mobile data.
- TOEFL iBT Score Scale Update: A Guide for Institutions — the 1–6 scale and its CEFR mapping, written for the universities reading your score.
Knowing what the test specifies is the easy half. The hard half is the errors that cost you marks on it, over and over, in the same places.
Why it works →