Celpe-Bras Exam Structure: Tasks, Levels, and Scoring

Celpe-Bras doesn't check grammar boxes on a test — it checks whether you can act in Portuguese: follow a news report by ear, write a letter to a newspaper, hold up your end of a conversation about a cartoon. The final score comes from two independent parts — Parte Escrita (written) and Parte Oral (oral) — and the certificate level isn't an average of the two: it's set by the lower of the two results. Below is what actually happens inside each part, where the níveis Sem certificação, Intermediário, and Avançado come from, and what an examiner is actually scoring in each task.

Parte Escrita: 4 tasks in 3 hours

The written part runs up to three hours and consists of four integrated tasks — tarefas. "Integrated" means each task first asks you to understand some input (watch a video, listen to a recording, read a text), then to produce your own text based on it: Celpe-Bras never tests reading or listening in a vacuum — those skills are always wired into the job of writing something of your own.

Every tarefa comes with a prompt (elemento provocador) and a specific communicative task: instead of "write about topic X," you're dropped into a situation — a blog reader, a neighbor, a forum member — and asked for a specific text, to a specific reader, for a specific purpose. The genre shifts from task to task: a formal or informal e-mail, a carta do leitor (a reader's letter to an editor), a cartaz-style notice, a forum or social comment, a note to colleagues. The timing works like this: the task 1 video is shown to everyone twice, and tarefa 1 gets 30 minutes including both viewings. Then the task 2 audio is played twice, and after that you divide the remaining two and a half hours across tarefas 2, 3 and 4 as you like — you can even go back to your first text. That’s a skill of its own: don’t get stuck on one text and leave the others with minutes to spare.

Tarefa 1 — from video to text

The first task is built on a short video: a news report, an interview, a segment. The clip is shown twice, and you write a text that follows logically from it — a letter, a comment, a note. A made-up example, just to show the logic (not from a real exam): you watch a report on plastic recycling in a city park and write a letter to a neighborhood newspaper proposing a similar initiative on your own street. A useful habit is jotting down details as you watch that you'll want to use later — memory alone is risky when time across all four tarefas is tight.

Tarefa 2 — from audio to text

The second tarefa follows the same logic minus the picture: the input is an audio recording, a radio segment, or an interview, also played twice. Following it by ear with no visual cue is harder, and this is the task that scares people most while preparing — even though the response is built the same way as in the first task. A made-up example: you listen to an interview with a volunteer at an animal shelter and write a blog post telling your followers about the initiative and inviting them to help.

Tarefas 3 and 4 — from print to text

The third and fourth tasks are built on printed texts — an article, a notice, a social post — but that doesn't mean they overlap: the genre of the input, the genre of the response, and the addressee all differ from one tarefa to the next. A made-up example for tarefa 3: you read a notice about the library's new hours and write a short announcement for your building's neighbors about the same change, in a different genre and for a different purpose than the source. A made-up example for tarefa 4: you read an opinion piece about paid downtown parking and write a comment on a neighborhood forum agreeing or disagreeing with it, backing your own view.

The idea running through all four tasks: the examiner isn't checking whether you "correctly" summarized the source — they're checking whether you hit the right genre, addressee, and purpose. Each text is read independently by two raters, who give it a single holistic score from 0 to 5; if their scores differ by more than one point, a third rater reads it. The task score is the average of the two, and the written score is the average of the four tasks.

The scale doesn’t describe separate “grammar points” — it describes how consistently the text sets up the communication the task asks for. At the top level, the text establishes the right relationship between writer and reader in the requested genre, carries out the requested action, uses the necessary information from the sources in its own words, and its language serves that genre. Lower down, things slip: part of the purpose is missed, passages read like a different genre, needed information is left out or distorted. A zero goes, for example, to a text on another topic, to a retelling of the source without any authorship of its own, or to a text that ignores the sources. That’s how the parameters are set out in the Documento base do exame Celpe-Bras (Inep, 2020).

How the scale changed

Many prep materials still describe the scoring as three criteria — that’s the earlier grid. Up to the 2013/1 edition it was split into three axes: adequação contextual (writer, reader, purpose, choice of information, medium), adequação discursiva (cohesion and coherence) and adequação linguística (vocabulary and grammar). In the 2013/2 and 2014/1 editions there were two axes: contextual and linguistic-discursive. Since the 2014/2 edition a single set of parameters has been used, with no axes. Even under the old grid, raters didn’t score each axis separately — the text got one score: Inep stresses that language was never assessed apart from how it serves the text.

The practical lesson hasn’t changed: strong grammar won’t rescue a text in the wrong genre, and a flawless formal letter where an informal social post was requested is still a weak answer, even with zero language errors.

That's the most common trap for first-timers: seeing a familiar topic in the elemento provocador, they write what they already know how to write — a ready-made "for and against" essay, say — instead of the requested genre. Examiners spot it immediately, and the text gets a low score no matter how clean the language is.

When checking your own text, it helps to ask three questions. They aren’t official criteria — just a way to break the holistic score into steps you can work on:

• The situation — did you correctly read who is writing, to whom, and why. A letter to a friend and a complaint to an administration about the same issue sound completely different: the greeting, the tone, the level of formality.
• The source and cohesion — which information from the input your reader needs, whether you restated it in your own words, and whether the text holds together: an opening and a closing, logical transitions between paragraphs.
• The language — do your words and grammar help you do the task? Errors matter most when they get in the reader’s way or break the genre: strong grammar in a text of the wrong genre still won’t earn a high score.

The format stays in-person and on paper: as of today there's no confirmed digital version of the exam, though that could of course change down the line.

Parte Oral: 20 minutes, two examiners

The oral part is shorter — about 20 minutes — and runs with two examiners present at once: the interlocutor carries the conversation with the candidate, while the observador stays out of it. Both of them score you. The format has two blocks.

• ~5 minutes — a warm-up conversation. The interlocutor asks about you: where you're from, what you do, why you're learning Portuguese — an unscripted chat, close to an ordinary first meeting.
• ~15 minutes — three elementos provocadores. The interlocutor then shows three prompts in turn, about five minutes each — usually photos, cartoons, short texts or charts — and asks you to comment on them, compare them, give your opinion. The point of this block is to hear connected, spontaneous speech, not a memorized monologue: describe what you see, connect it to your own experience or opinion, and react to a follow-up question — a strategy that works no matter which of the three prompts you get.

The two examiners use different grids. The interlocutor gives one overall, holistic score. The observador uses an analytic grid with six aspects of your speech — listening comprehension, interactional competence, fluency, vocabulary, grammar and pronunciation — weighted differently. Both score on a 0–5 scale, and the oral score is the average of the two. If the examiners differ by 1.5 points or more, or the oral score is more than 2 points below the written one, the recording is listened to and scored again.

As in the written part, there's no single correct answer being checked — what's scored is how freely and appropriately you develop speech around the given prompt: keeping the thread going, reacting to what the interlocutor says, rather than reciting a prepared text that barely connects to what was shown or asked.

Levels and scores

The overall Celpe-Bras scale runs from 0 to 5, split into five bands (Documento base, Quadro 10):

Note: below Intermediário no certificate is issued at all — Sem certificação means the attempt simply didn't pass, not that you got a "weak" certificate. Unlike some language tests, which hand out a scaled score regardless of result, Celpe-Bras has a hard entry threshold, and preparation should aim at clearing it, not at landing "somewhere" on the scale.

Intermediário is the minimum level a certificate is actually issued for. It is often compared to B1, but Inep doesn’t publish an official CEFR equivalence. For naturalization, a Celpe-Bras certificate is one of the recognized ways to prove your Portuguese, not the only one.

How the final result is computed: the written score is the average across all four tarefas; the oral score is the average of the two examiners' scores (interlocutor and observador). To get a certificate you need at least Intermediário in both parts, and if the two parts land on different levels, the lower one counts — there's no compensation between them. The rule is stated directly in the Documento base.

A made-up example to make the rule concrete (numbers invented, not from a real exam): the written part averages 3.80 across the four tarefas — already Avançado territory. The oral part scores 2.40 — Intermediário territory. The certificate is issued at the lower score, Intermediário, not Avançado: a strong written part doesn't pull a weak oral part up with it.

The practical upshot isn't always obvious at the start of preparation: if you read well and know plenty of vocabulary but have never once written under a clock or spoken out loud with a conversation partner, either the written or the oral part can drag the result down regardless. Preparation needs to cover every type of activity at once, not just the one that already feels comfortable — leaning only on grammar or reading is the most common reason for a lower-than-expected result.

What this means for preparation

The exam's structure is, in effect, a ready-made study plan: it scores four kinds of writing plus live spoken interaction, so that's exactly what needs practicing — not whatever happened to be in a textbook. Three takeaways follow.

• Practice writing genres against a clock, not just isolated grammar points: a reader's letter, a notice, and a social comment differ in tone, structure, and how you address the reader — something you need to feel in your hands, not just know in theory. It sticks best in small, regular doses rather than crammed into the last month.
• Listen to and watch authentic audio and video, not only simplified textbook dialogues — Tarefa 1 and 2 are built on real Brazilian speech at natural speed, with real pauses, slang, and regional accents, not studio narration written for a course.
• Speak out loud regularly, including about random photos and cartoons — Parte Oral doesn't forgive silence or "let me think" pauses: what's valued is being ready to develop an idea immediately, even when the topic catches you off guard.

None of these three skills comes from a textbook — only from practicing under exam-like conditions: against a clock, out loud, on unfamiliar material.

Mnezo's dedicated Celpe-Bras prep mode is built around exactly this: enter the date of your upcoming session, and the app tracks readiness as a percentage across separate dimensions — words, writing, reading, listening, grammar — and flags what to shore up first, following the same "the weakest link drags the total down" principle covered above.

A good source for practicing on real tasks is the archive of past exams; for where to find it and how to use it, see "Celpe-Bras past papers: how to practice on real tasks". And if you'd rather live through the full Parte Escrita format yourself than just study other people's tarefas — 4 tasks, 180 minutes, with a full review — Mnezo has an AI-graded Celpe-Bras written mock exam; more on it in "Celpe-Bras practice test: how the mock exam works".

The short version: Celpe-Bras is four writing tasks in three hours plus a live conversation with two examiners, scoring not grammar or vocabulary in isolation but the ability to hit the right genre, addressee, and purpose, in writing and in speech alike. The final level is dragged down by the weaker half, so even preparation across every skill beats leaning hard on one strong one.