The Most Dangerous Roads in Brazil: What the Data Can and Cannot Say

The ranking exists, and it measures the wrong thing

Brazil's federal highway police publish their crash records as open data, one row per occurrence, and the file names the road: a column for the BR number, a column for the kilometre marker, plus latitude and longitude. Those fields are filled in on all 72,529 records for 2025, which is why the ranking can be built at all.

Reading that file for 2025 gives the order everyone quotes: BR-101 with 760 deaths across 13,014 occurrences, BR-116 with 708 across 11,021, then BR-153 with 282 and BR-040 with 214.

The problem is what is missing. There is no traffic count anywhere in the dataset — no vehicle-kilometres, no average daily flow. Dividing deaths by exposure is what turns a count into a risk, and that division is not possible with what is published. The BR-101 and BR-116 lead the list because they are the longest and busiest roads in the country.

🔴 So the honest sentence is "the BR-101 is where most people die", and the dishonest one is "the BR-101 is Brazil's most dangerous road". They are not the same claim, and only the first has a source.

⚠️ There is a second limit, and it is bigger than the first. That file covers federal highways only — 6,043 of the 21,376 road deaths Brazil recorded in 2025, or 28%. Roughly three in four road deaths in Brazil happen somewhere this dataset does not look, and the national register that does cover every road has no field naming the highway at all.

The number that should change how you drive

Strip out the ranking and the same file answers a better question: what kind of road kills.

An undivided highway takes 48% of the crashes and 69% of the deaths. Per crash it is 2.3 times more lethal than a dual carriageway — because a head-on collision is what an undivided road makes possible and a barrier does not.

This matters more to a visitor than to a Brazilian, for one reason: in most of the countries our readers drive in, "highway" means a divided road by definition. In Brazil a BR-numbered federal highway is very often one lane each way, with a painted line between you and a lorry. Anyone planning a long drive should read that table before choosing a route.

Two more cuts from the same file, both counter-intuitive. Night kills more than day — 2,892 deaths in full darkness against 2,391 in full daylight, on far less traffic. And three quarters of the deaths are on rural stretches: 4,509 rural against 1,534 urban.

Two official databases, opposite answers

This is the finding that should make anyone careful with a Brazilian road-safety statistic, whichever way it points.

Between 2022 and 2025, deaths on federal highways as measured by the highway police rose 11.1% — 5,441 to 6,043. Over exactly the same years, deaths on all Brazilian roads as measured by the national register fell 4.9% — 22,480 to 21,376.

Both are official. Neither is wrong. They measure different roads, and choosing one gives you the opposite headline from the other. Any article that reports a trend without saying which register it used has not told you enough to check it.

There is a further limit that no amount of care gets around: neither database records the date of death. That is not a poorly documented definition — the field does not exist. So the Brazilian figures cannot be reconciled with the international thirty-day convention, and comparing them directly with a European or American road-death rate compares two different things.

📌 And the study the press calls "Brazil's worst roads" is not about crashes at all. It is a survey of pavement, signage and geometry — the condition of the surface — published by a private transport-industry confederation, not by government. It is a real and useful survey. It is simply not a measure of danger, and it gets cited as one.

"Road" and "highway" are different words in Brazilian law

A definition in the traffic code catches out drivers who assume the two are synonyms, because it changes the speed limit.

The code defines a rodovia as a paved rural road and an estrada as an unpaved rural road. Where no sign says otherwise, the default limit on a rodovia is 100 km/h and on an estrada 60 km/h. The distinction is the asphalt, not the size — and the sign you did not see is not a defence.

This is one of several rules that differ from what visitors assume, and they are the ones worth knowing before you turn the key rather than after. Driving in Brazil as a foreigner covers the licence side, specifically with an American licence, with a British one, and whether the international permit is required.

The rules that changed, and the ones that never move

Headlights. Almost every English-language guide still repeats the 2016 rule that daytime headlights are compulsory on highways. A 2020 law rewrote it: daytime low beam on an undivided rural highway is now required only of vehicles without daytime running lights. What did not change, and applies to everyone: low beam is compulsory at night, in tunnels, and in rain, fog or mist — even at midday.

Alcohol. Brazil does not have a limit that you can drink up to. The traffic code penalises any concentration of alcohol, and the penalty for driving under the influence is the heaviest grade of infraction: a fine multiplied by ten, twelve months' suspension of the right to drive, and the vehicle retained.

Children. A child under 10 who has not reached 1.45 m travels in the back seat, in a restraint device. That is stricter than in several of the countries our readers come from.

Who to call. There are two numbers on the road, not one: 191 for the federal highway police on a BR-numbered road, 198 for the state highway police. Both are on the regulator's list of free emergency services. The 166 you may see advertised belongs to the federal transport regulator, sits on the other list — ordinary public utility, chargeable as a local call — and is a complaints channel, not rescue. On a tolled federal highway there is no single national rescue number: each concession runs its own line, and the number is on the signs along that road. Which emergency number does what is worth reading before the trip.

If the car is not yours, two more things follow: what the rental protection quietly excludes, and what compulsory vehicle insurance in Brazil actually is, which is not what most visitors assume. On the same drive, tolls now read your plate and the hold on your card is the part that ruins budgets.

Sources

• Federal Highway Police, open data — the 2025 crash file grouped by occurrence, from which we computed the totals, the ranking by BR number, the road-type table, the day-phase split and the rural/urban split. Accessed September 8, 2026.
• RENAEST, national road accident register — the national totals for 2018 to 2025, read in full from the published tables. Accessed September 8, 2026.
• Brazilian Traffic Code — Annex I on rodovia and estrada, art. 40 on headlights as amended in 2020, the alcohol provisions and the rules on child restraint and seat belts. Accessed September 8, 2026.
• Anatel, public utility and emergency codes — 191 and 198 as free emergency services, and 166 on the separate, chargeable list. Accessed September 8, 2026.
• CNT highway survey — the study cited abroad as a ranking of dangerous roads, which measures pavement, signage and geometry, and is published by a private industry body. Accessed September 8, 2026.

What we could not verify

• Which Brazilian highway is most dangerous per vehicle travelled. No official source publishes a traffic denominator alongside the crash data, so the ranking cannot be converted from volume into risk. This is the central question of the subject and it has no published answer.
• Any comparison between a state highway and a federal one. The national register has no field naming the road, and its road-description fields are empty in 94% to 98% of its 8.5 million records. There is no basis for the comparison in either direction.
• How Brazilian road-death figures compare with countries using the thirty-day rule. Neither Brazilian database records when the victim died, so the adjustment cannot be made at all.
• How much under-reporting there is. Quantifying it would require the national mortality system, which we did not consult, so we state neither that the figures are complete nor that they are undercounts.
• What each crash was actually caused by. The cause field in the police file is the attending officer's judgement at the scene, not a forensic finding, and the official variable dictionary for that file dates from 2017 and contradicts the data it describes in three of four checkable specifications.