Is Brazil Safe to Travel? What the Numbers Actually Say

Two national numbers, both correct

For 2025 the Ministry of Justice published 31,448 intentional homicides, a rate of 14.74 per 100,000. The Brazilian Public Security Forum published 40,775 violent deaths, a rate of 19.1. Same country, same year, a 30% gap.

Neither is wrong. The federal figure counts intentional homicide and excludes by name femicide, bodily harm followed by death, robbery followed by death, and deaths caused by police intervention. The Forum's headline number — morte violenta intencional — adds those categories back in. Add the four excluded parcels to the federal figure and you land within 0.6% of the Forum's total, which is what confirms the two are describing the same reality through different definitions.

Two consequences worth carrying. A journalist quoting 19.1 and one quoting 14.74 are both citing an official source — so a contradiction between two articles is not necessarily one of them being sloppy. And neither organisation collects the data: both consolidate what each state hands over, which is where the revisions and the inconsistencies enter.

The national average describes nowhere

State rates in 2025 ran from about 7.6 to over 40 per 100,000 — a spread of more than five times. The Northeast alone accounted for 46.7% of the country's homicides.

City-level data breaks the average further, and it contains a warning for travellers. Among the ten cities with the highest violent-death rates is Porto Seguro, a beach destination in Bahia, at 71.2 per 100,000. But the Anuário notes that a large share of those deaths — close to half — are deaths from police intervention. Quoting the rate without that composition tells a visitor something the number does not actually say.

Homicide in Brazil is also overwhelmingly a male phenomenon, and concentrated in ways that have little to do with tourist itineraries. That is not reassurance; it is a reason the national rate is a poor instrument for a personal decision.

Four governments, four incompatible maps

Read the four English-language travel advisories side by side and they do not describe the same country.

The United States places Brazil at level 2 of 4, with pockets at level 4 — including a 100-mile strip along several land borders, all favelas, and Brasília's satellite cities at night. It also tells its own staff not to use municipal buses in some cities. The United Kingdom does not use levels: it advises against travel to a handful of specific river stretches in the Amazon and little else. Canada names individual neighbourhoods, putting two Rio complexes in its highest category. Australia sets a nationwide caution and draws no internal lines at all.

The single convergence is the useful part, and it runs against a visible tourism industry: all four say not to enter a favela, guided tour included.

One caveat on how you read these. Their update dates are not comparable — the American advisory a traveller reads today has been unchanged since May 2025, while the Canadian page was revised this week. An advisory is a document with an age, and the age is rarely on the part you read.

What the numbers do not cover

Ordinary crime — pickpocketing, phone snatching, the thing most likely to actually happen to you — is measured far worse than homicide, for a simple reason: it has to be reported first.

Brazil measured that gap. A national victimisation survey run with the statistics institute for 2021 found that, for incidents outside the home, victims went to the police in 44.8% of thefts and 57.9% of robberies. So roughly half of ordinary property crime never enters any statistic, in either of the two national datasets above.

Brazil has no periodic victimisation survey — these exercises happen years apart — which means the correction factor itself is dated between rounds.

And there is one gap that matters specifically here: no Brazilian open dataset separates victims by nationality. Rio does run a police station dedicated to crimes against tourists, so the information exists administratively; what does not exist is a published series. Anyone quoting a "crime against tourists in Brazil" rate is extrapolating.

Rio measures at street level, and the two risks are inverted

Where the national picture is coarse, Rio de Janeiro publishes an open base by police-station area, monthly, which makes it possible to compute rates for a neighbourhood rather than a country.

Do that for 2025 and something counterintuitive appears. The station area covering Ipanema and Leblon — the classic tourist strip — recorded a homicide rate of about 3 per 100,000, against 22.6 for the state. Meanwhile the same district sits near the top of the city for phone crime.

The two risks run in opposite directions. In the tourist zone the phone crime is overwhelmingly theft — taken without confrontation — while in outer districts the same category is dominated by robbery, taken with force. Two different events wearing one label in the statistics.

Two honest limits on this. The population denominators come from the 2022 census while the crimes are from 2025, which distorts districts with small residential populations and large daytime crowds. And any single year is noisy: a ratio between two districts that looks dramatic in 2025 is much smaller averaged across recent years.

São Paulo publishes something even more granular for phone crime — street-level records with coordinates — but the state's own methodology page warns that the file is not the official statistic and that a large share of entries are supplementary filings, so the row count is not an incident count.

Emergency numbers, and why there is no 911

190 for police, 192 for medical emergencies, 193 for fire. There is no single number.

The structural reason is worth knowing, because it explains why the situation is unlikely to change soon: the three-digit emergency codes are not one national system. Each is contracted locally by the relevant authority, which is why coverage, response and language capability vary by city rather than by country. Do not assume an English-speaking operator.

Also worth knowing before you travel: at least one foreign government's own advice page for Brazil tells readers to dial 911, which does not work here — boilerplate that survived a revision of the very section it sits in.

Practical adjacent reading: ride apps and how they are regulated here, getting around São Paulo on public transport, and the ride in from the airport, which is the leg where most first-time visitors feel most exposed. On money, carrying less is the simplest mitigation — how to get cash without a local card, which ATM to use, and how much you can legally bring in. Before any of that, whether you need a visa and getting a local SIM, because a working phone is the thing every safety instruction quietly assumes. And the other question people ask in the same breath: whether you can drink the tap water.

Sources

• Anuário Brasileiro de Segurança Pública 2026 (FBSP) — the 19.1 figure, state and city breakdowns, and the composition note for Porto Seguro. Accessed September 3, 2026.
• Mapa de Segurança Pública 2026 (Ministry of Justice) — the 14.74 figure and the written definition excluding four categories. Accessed September 3, 2026.
• PNAD Contínua victimisation survey — reporting rates of 44.8% for theft and 57.9% for robbery. Accessed September 3, 2026.
• ISP-RJ open data and its methodology page — the police-station base used for the neighbourhood rates, and the counting rules. Accessed September 3, 2026.
• Travel advisories: US State Department, UK FCDO, Government of Canada and Smartraveller. Accessed September 3, 2026.
• Anatel, public utility and emergency numbers — the structure behind the three-digit codes. Accessed September 3, 2026.

What we could not verify

• Any rate for crime against foreign visitors. No open Brazilian base carries nationality as a variable. The administrative data exists — Rio has a dedicated station — but is not published as a series.
• How much of the tourist-district phone-crime figure involves visitors rather than residents. The base records the offence and the location, not who the victim was.
• Official case numbers for the 2026 methanol poisoning alerts, which appear in more than one foreign advisory. The Brazilian source pages we reached returned a restricted-content page rather than the data.
• Whether the police-intervention share behind Porto Seguro's rate holds for the most recent year. The composition note is published for the series, not year by year.