Violent Crime Is Down About 60% Since 1991. That's Roughly 10,000 Americans Not Killed Each Year — And We Still Don't Know Why.

The violent crime rate in the United States is roughly 60% below where it peaked in 1991. Translate that into bodies and it's about 10,000 to 11,000 fewer Americans killed every year — the population of a small town, saved annually, for a generation running. Almost nobody is celebrating. And almost nobody can fully explain why.

The 60% Number

“Crime is down 60%” is the kind of headline that's easy to wave off as cherry-picked. It isn't. The framing is sturdy across multiple datasets, multiple definitions of crime, and three decades of methodological scrutiny.

• The FBI's Uniform Crime Reports show a violent crime rate of roughly 758 per 100,000 in 1991. By 2024 it sat near 364 per 100,000 — about a 52% drop in the official measure, with 2025 estimates running meaningfully lower.
• The Bureau of Justice Statistics' National Crime Victimization Survey — which asks households whether they were victimized, rather than relying on police reports — shows violent victimization down more than 70% over the same window.
• Property crime has fallen even more consistently: about 60% over the last 30 years, with the trend line less jagged than violent crime's.

Pick whichever dataset you trust most and the answer rounds to the same place: violent crime in America is somewhere between half and two-thirds lower than it was at its modern peak. The 60% figure is the rough midpoint, and it's honest.

Now Translate It Into Bodies

Crime rates abstract people away. Homicide is one of the only crime categories where the count itself is reasonably comparable across decades — agencies can disagree about what constitutes an aggravated assault, but a homicide is a body. So we'll work in bodies.

In 1991, the United States recorded roughly 24,700 murders. In 2025, the count was approximately 14,000 — the lowest figure in decades and possibly the lowest rate since the early 1960s. The arithmetic of that difference is:

Some context for the scale of 10,700 lives a year:

• That's more than the U.S. lost to drunk driving in 2023 (about 13,500), and not far below the toll of firearms suicide in a given year.
• It's roughly the population of a small American town — places like Forest Hills, NY or Hanover, NH — saved from the homicide column every year.
• Compared to other public-health wins, it's on the same order of magnitude as the lives saved annually by seatbelt laws. We didn't skip the ribbon-cutting on seatbelts.

And Yet — Two Truths

This is where a serious crime-data publication has to hold two truths at the same time. The national arc is real. So is the experience of someone living on a block where the homicide rate is still 30 per 100,000 — triple the national 1991 peak — and where the “60% down” story sounds like a sick joke.

National averages mask enormous local variation. A small number of cities, and inside those cities, a small number of census tracts, account for an outsized share of remaining violence. The U.S. as a whole getting safer doesn't mean every block is. SpotCrime's own data — incidents reported at the block level, updated continuously — shows scores swing dramatically across distances measured in blocks, not miles.

So both of these are true:

• America is dramatically safer than it was a generation ago.About 10,000 Americans who would have died from homicide at 1991 rates are alive each year. That's not a press release; that's arithmetic.
• Hyperlocal risk still matters intensely.Telling someone in a high-incident block that the national rate is down doesn't make their commute safer. National trends and block-level reality live at different scales.

Both pieces matter. The first is why the policy conversation should be calmer than it is. The second is why every piece of address-level crime infrastructure SpotCrime builds matters more, not less, in a national downtrend.

The Hard Part: We Still Don't Fully Know Why

Crime declines are over-explained after the fact and under-predicted before. The 1990s decline had at least a dozen plausible mechanisms — Roe v. Wade, lead abatement, the crack epidemic burning itself out, broken windows, mass incarceration, demographics, economic boom, immigration — and serious researchers still disagree about the relative weights three decades later.

The 2023–2025 decline is going to be the same conversation. The pandemic spike of 2020 reversed faster and harder than almost anyone predicted: 2023 set a record for the largest one-year decrease in murder, 2024 broke it, and 2025 broke it again. Candidate explanations include the unwinding of pandemic-era social disruption, the maturation of community violence intervention programs, demographic shifts, smartphone-driven indoor activity displacing street time, and — uncomfortably for some — the fading of the post-2020 protest cycle. None of these is conclusive on its own.

We covered the 2024–25 chapter of this in detail in our companion post, Why Has Violence Fallen Off a Cliff?. The longer historical arc — and why the public hasn't caught up to it — is covered in America's Historic Crime Drop. And we follow up on the institutional question — why almost no one is funding the research that would tell us why this is happening — in part two of this series, If Cancer Had Dropped Like This, We'd Know Why.

What's worth saying here is the meta-point: a decline whose mechanism we don't fully understand is a decline that could reverse. Lead abatement isn't going to un-happen. But pandemic-related disruption can recur; community violence interruption programs are politically fragile; the next economic shock will be a real-time experiment in which of the proposed mechanisms actually held the line. Watching block-level incident data is how you see a reversal before the national numbers print.

Why This Matters for Anyone Using Crime Data Operationally

If you build a real estate platform, an executive protection program, a family-safety app, or a corporate security operation that consumes crime data, the 60% decline reframes what your tools should actually do.

• Calibrate for current reality, not 1990s priors. Risk scoring built on assumptions from two decades ago will systematically overweight neighborhoods that have changed dramatically. Refresh your baselines.
• Hyperlocal granularity is the differentiator now.When the national trend is down, the valuable signal isn't whether a city is risky — it's which block in that city is. A ZIP-code average is a worse approximation of reality today than it was in 1991, because the within-ZIP variance has grown relative to the between-city variance.
• Watch for reversal signals locally.National crime statistics print on a 6-to-18-month lag. Block-level reported incidents are the early-warning system. A risk product that only updates with the FBI's release cycle will miss the turn — in either direction.

A Boring Conclusion to a Remarkable Story

The most important crime statistic in America right now is the one we're collectively underreacting to: there are roughly 10,000 people alive at the end of each year who, at 1991 rates, would not have been. They didn't make the news. They're people who didn't have a worst day, didn't get a knock at the door, didn't become a number in a press release. The non-event is the story.

Honoring that story means two things at once. Saying clearly that the country is safer than the cultural mood reflects. And refusing to flatten the people who still live where the decline hasn't reached yet. That's the calibration SpotCrime's data is built to support — at the block, the address, and the point.