Every U.S. civil aircraft that crashes, catches fire, or breaks apart in flight is, in the end, a file: a date, a make and model, a phase of flight, and a probable cause the National Transportation Safety Board spent months reconstructing. Most of the file is small airplanes on quiet fields. A handful are the crashes that rewrote the rules. This is a reading of what actually brings airplanes down, where in the flight it happens, and the two very different worlds of commercial and private aviation.
Read across four decades: U.S. aviation accidents have fallen by roughly 66% from their early-1980s peak, and the fatal ones fell with them. Even so, about one accident in five still ends in death - the red line never closes on zero. Nearly all of this is general aviation; the scheduled airlines are a rounding error at this scale.
Real - counted from CAROL Every U.S. aviation accident by year of occurrence, 1982-2025, with the fatal subset. The current year is held off (partial). See Methodology.
In 2025, about 1,156 U.S. aviation accidents, 166 of them fatal (14%). The peak was 3,402 in 1982. Counts are of occurrence year; the record is comprehensive from 1982.
| Year | Accidents | Fatal | Fatal % |
|---|---|---|---|
| 1982 | 3,402 | 627 | 18% |
| 1983 | 3,352 | 645 | 19% |
| 1984 | 3,248 | 603 | 19% |
| 1985 | 2,932 | 543 | 19% |
| 1986 | 2,718 | 497 | 18% |
| 1987 | 2,623 | 476 | 18% |
| 1988 | 2,505 | 483 | 19% |
| 1989 | 2,377 | 466 | 20% |
| 1990 | 2,366 | 485 | 20% |
| 1991 | 2,309 | 462 | 20% |
| 1992 | 2,212 | 472 | 21% |
| 1993 | 2,147 | 418 | 19% |
| 1994 | 2,108 | 425 | 20% |
| 1995 | 2,126 | 422 | 20% |
| 1996 | 2,013 | 374 | 19% |
| 1997 | 1,945 | 360 | 19% |
| 1998 | 2,007 | 371 | 18% |
| 1999 | 2,000 | 344 | 17% |
| 2000 | 1,951 | 363 | 19% |
| 2001 | 1,832 | 345 | 19% |
| 2002 | 1,815 | 357 | 20% |
| 2003 | 1,869 | 369 | 20% |
| 2004 | 1,722 | 342 | 20% |
| 2005 | 1,781 | 329 | 18% |
| 2006 | 1,596 | 316 | 20% |
| 2007 | 1,742 | 298 | 17% |
| 2008 | 1,641 | 283 | 17% |
| 2009 | 1,549 | 268 | 17% |
| 2010 | 1,494 | 265 | 18% |
| 2011 | 1,550 | 278 | 18% |
| 2012 | 1,533 | 270 | 18% |
| 2013 | 1,282 | 220 | 17% |
| 2014 | 1,281 | 253 | 20% |
| 2015 | 1,288 | 234 | 18% |
| 2016 | 1,318 | 210 | 16% |
| 2017 | 1,310 | 205 | 16% |
| 2018 | 1,337 | 224 | 17% |
| 2019 | 1,296 | 240 | 19% |
| 2020 | 1,132 | 203 | 18% |
| 2021 | 1,207 | 214 | 18% |
| 2022 | 1,260 | 204 | 16% |
| 2023 | 1,194 | 195 | 16% |
| 2024 | 1,174 | 181 | 15% |
| 2025 | 1,156 | 166 | 14% |
What the wreckage says, ranked. These are the leading probable-cause categories for fatal general-aviation accidents - the world where nearly all the crashes are. One pattern dominates every year: the pilot loses control of a perfectly good airplane. Weather and low-altitude maneuvering kill a smaller share, but far out of proportion to how often they happen.
Real - NTSB defining-occurrence coding Counted from the NTSB bulk accident database: the CICTT occurrence the NTSB coded as defining for each fatal Part 91 accident (1,960 over 2014-2023), crosswalked into these cause families. A cause is flagged high fatality when accidents in it turn fatal at over 1.25x the general-aviation average. See Methodology.
Stalls, spins, and upsets the pilot never recovers from - the single largest killer in general aviation, nearly half of every fatal crash.
Engine and system failures. A small share of all accidents, but when the engine quits over bad terrain the forced landing turns fatal - the second-largest fatal category.
Midair collisions, birdstrikes, ground mishaps, and the cases where the wreckage never gives up a single cause.
A flyable aircraft flown into ground, water, or obstacles - most often at night or in reduced visibility.
A visual pilot continuing into cloud or losing the horizon. When it is the defining event, it is fatal more often than any other cause.
Running the tanks dry or feeding from the wrong one - the most preventable line in the file.
Buzzing, canyon-turns, low aerobatics: little room, no margin. Rare as a defining event, lethal when it is one.
share of fatal GA accidents · disproportionately fatal when it occurs
Read it as the arc of a flight, top to bottom: taxi out, climb, cruise, and come back down. Each bar is the share of accidents that involved that phase - and because a single accident can span several (an engine quits enroute, the forced landing bends metal, a post-impact fire follows), the shares overlap and add to more than 100%. Two ends dominate: landing, where the airplane is low, slow, and changing configuration, is in nearly half of all accidents - and is among the least fatal. Takeoff, maneuvering, and the phases tagged fatal-heavy are where the deaths concentrate.
Real - counted from CAROL Counted from the NTSB CAROL query API (EventSequence phase groups), 2014-2023. Bar length is the share of accidents that involved each phase; because one accident can span several phases, the shares overlap rather than partition. The phases tagged ▲ fatal-heavy are those whose accidents were fatal at a disproportionately high rate - well above the overall average. See Methodology.
Landing is involved in about 48% of accidents - the single most common phase, and among the least likely to be fatal. Because one accident can involve several phases, the shares overlap and do not sum to 100%.
| Phase | Accidents involving | Fatal-heavy |
|---|---|---|
| Standing | 7% | - |
| Pushback / tow | 0% | - |
| Taxi | 3% | - |
| Takeoff | 13% | - |
| Initial climb | 10% | yes |
| Enroute | 18% | yes |
| Maneuvering | 14% | yes |
| Approach | 14% | yes |
| Landing | 48% | - |
| Emergency descent | 12% | - |
| Uncontrolled descent | 9% | yes |
| Post-impact | 2% | yes |
| After landing | 0% | - |
| Unknown | 1% | yes |
The same sky holds two almost unrelated safety records. Scheduled airline flying has become the safest way humans have ever moved: whole years pass without a single passenger death. Private flying carries nearly the entire accident burden - a steady drumbeat that has barely changed in decades. Every figure here is real: the counts from the NTSB record, and the rates against real flight-hours - general aviation from the FAA activity survey, the airlines from the NTSB's air-carrier statistics.
The safest way humans have ever moved. Between Colgan Air 3407 (2009) and the Reagan National midair (2025), no U.S. airliner killed a scheduled passenger for roughly sixteen years.
Private flying carries almost the entire U.S. accident burden - a steady drumbeat of roughly a thousand crashes and a couple hundred fatal ones every year, largely unchanged for decades.
Major scheduled-passenger (Part 121) accidents with fatalities. The long quiet after 2009 is the record itself: no U.S. airliner killed a scheduled passenger for roughly sixteen years, until the Reagan National midair.
Engine and pylon separated on takeoff; maintenance-induced.
Took off with ice on the wings; engine anti-ice off.
Took off with flaps and slats not set.
Uncommanded rudder hardover; 737 rudder design.
Cargo fire from mislabeled oxygen generators.
Center wing fuel tank exploded.
Jackscrew failure; maintenance lubrication.
Vertical stabilizer separated after rudder inputs.
Took off from the wrong, too-short runway.
Stall on approach; crew's response and fatigue.
Midair with an Army helicopter near Reagan National. NTSB investigation ongoing.
A handful of accidents changed how everyone flies. Each of these has a published NTSB probable cause and a legacy written into regulation, training, or design. One of them - the Hudson ditching - is where this file meets its sister database of wildlife strikes. Findings below are paraphrased from the NTSB reports.
Probable cause The left engine and its pylon tore away during takeoff rotation, damaged the wing's leading edge, and retracted the outboard slats. Asymmetric lift rolled the DC-10 as it stalled. The NTSB traced the damaged pylon to a time-saving maintenance procedure that lifted the engine and pylon as one unit.
Legacy Still the deadliest aviation accident on U.S. soil. It forced a rethink of engine-change procedures and of stall-warning systems that went dark when the crew needed them most.
Probable cause The crew took off in a snowstorm without turning on engine anti-ice, so iced-over probes fed them false, high thrust readings. The undersped 737 never climbed, struck the 14th Street Bridge, and fell into the frozen Potomac. Ice on the wings and a decision not to reject the takeoff sealed it.
Legacy A textbook cold-weather and crew-coordination case. It reshaped de-icing standards and how crews challenge a takeoff that feels wrong.
Probable cause The tail-mounted engine's fan disk broke apart from an undetected fatigue crack in the titanium. Shrapnel severed all three hydraulic systems and left the crew with no flight controls at all. They flew the crippled jet to Sioux City on engine thrust alone.
Legacy A survivable outcome from an unsurvivable failure - 184 lived. It drove changes to titanium inspection and to how crews improvise with total control loss.
Probable cause Chemical oxygen generators, improperly labeled and packed by a maintenance contractor, ignited in the forward cargo hold. Fire filled the cabin and the DC-9 dived into the swamp minutes after takeoff.
Legacy It made fire detection and suppression mandatory in Class D cargo holds and hardened the rules for shipping hazardous materials by air.
Probable cause The nearly empty center wing fuel tank held a flammable fuel-air vapor. A short circuit outside the tank most likely carried energy into it and set it off, blowing the 747 apart off Long Island.
Legacy One of the most exhaustive investigations ever run. It produced fuel-tank flammability rules and the inerting systems now standard on transport jets.
Probable cause On an icy night approach to Buffalo, the stall warning fired and the captain pulled back instead of pushing forward, driving the Q400 into a full stall it never recovered from. The NTSB cited the crew's response, fatigue, and lax cockpit discipline.
Legacy The last U.S. airline crash to kill scheduled passengers for roughly sixteen years. It produced the 1,500-hour first-officer rule and sweeping pilot-fatigue and training reforms.
Probable cause A flock of Canada geese went into both engines seconds after takeoff from LaGuardia, killing all thrust. With no runway in reach, the crew ditched in the Hudson. Everyone aboard survived.
Legacy The 'Miracle on the Hudson' - and the point where this file touches its sister database of wildlife strikes. It sharpened bird-strike research and ditching and engine-restart procedures.
Probable cause On a clear-day visual approach, the crew let the airspeed decay while relying on autothrottle behavior they misunderstood. The 777 struck the seawall short of the runway. The NTSB cited mismanagement of the approach and over-reliance on automation.
Legacy A defining automation-dependency case. It pushed carriers toward more hand-flying proficiency and clearer autothrottle training.
Every figure here is now measured from the real record - the NTSB investigation database, its coded bulk file, and official flight-hours from the FAA and NTSB. This page documents where each number comes from.
The authoritative record is the NTSB aviation investigation database, searchable through CAROL (Case Analysis and Reporting Online), with individual accident reports published at the same site. It reaches back to 1962 and holds tens of thousands of investigated accidents. The cohort cadences on this page are counted straight from CAROL's query API (exact accident and fatal tallies per operating rule); the individual case files - the Hall and the airline timeline - are hand-assembled from published NTSB reports.
The Hall of Investigations and the airline-fatal-accident timeline are real. Every case - date, aircraft, location, fatality and survivor counts, and the probable cause - is drawn from the published NTSB report for that accident. Probable-cause text is paraphrased for length, not quoted verbatim. The cohort cadences in "Two Industries" and the compare tool - accidents per year and fatal-accident cadence for Part 121, 91, and 135 - are exact counts from CAROL over the most recent full decade (2014-2023), which is the honest "current" rate: U.S. aviation accidents have fallen so sharply since the 1970s-80s that an all-time average would badly overstate today. The record bears out the contrast: scheduled airlines logged just four fatal Part 121 accidents across that decade - years apart - while general aviation ran to roughly 1,100 accidents a year.
The phase-of-flight profile is now real too, counted from CAROL's EventSequence phase groups over the same 2014-2023 window. One caveat governs how to read it: a single accident's sequence can touch several phases, so the counts are phase involvement, not a mutually-exclusive split - the shares overlap and add to more than 100%. Each bar is the share of accidents that involved that phase, and a phase is tagged "fatal-heavy" when accidents involving it were fatal at a disproportionately high rate - at least 1.25 times the overall average. Landing is involved in nearly half of all accidents and is among the least fatal; the deaths concentrate in maneuvering, takeoff, and the uncontrolled phases.
The probable-cause shares were the site's last estimate; they are now Real. CAROL's query API carries no countable cause category, but the NTSB bulk accident database does: every accident's sequence of events is coded, and one occurrence is flagged as defining. We count that defining occurrence - a CICTT category, not free text - for each fatal Part 91 accident over 2014-2023 and crosswalk the categories into the seven families shown. The same coding fills the Part 135 cause chart in the compare tool (its fatal record); the airline cause chart there stays the landmark-case corpus. A family is flagged high fatality when its accidents turn fatal at more than 1.25 times the general-aviation average - itself a real figure, computed from the coded record (fatal versus all Part 91 accidents in that family).
The NTSB names a single probable cause for each accident, and codes a full sequence of events with one occurrence marked defining. The categories here collapse that nuance into one bucket per accident. A stall that followed an engine failure that followed fuel starvation is one story with several honest labels; this page counts the single occurrence the NTSB coded as defining.
A rate needs flight-hours exposure. The general-aviation and Part 135 rates use the FAA's General Aviation and Part 135 Activity Survey (official survey-based hour estimates): about 5.2 accidents per 100,000 hours in GA (0.9 fatal), 1.1 in Part 135. The Part 121 (airline) rate - about 0.16 per 100,000 hours, 0.0023 fatal - uses the NTSB's Aviation Accident Statistics air-carrier file, whose flight-hours are FAA-compiled.
One honest wrinkle there: the NTSB's air-carrier file counts 282 Part 121 accidents over 2014-2023, because it includes U.S. carriers' international operations, while the accident count shown in the cards (257) is the CAROL query's U.S.-only tally. For the rate we deliberately pair the NTSB file's own numerator and denominator (282 accidents against its 176.6 million flight hours) so that top and bottom share the same "14 CFR 121, all operations" scope. Dividing the U.S.-only 257 by the all-operations hours would understate the rate. The fatal count is 4 either way.
Incidents that never rose to a reportable accident. The near-misses. The maintenance discrepancy caught on the ground. What ships here is the investigated-accident record, its causes, and the rates that record produces against real activity.
Feeding scripts/build-data.ts: fetch-carol.mjs counts the cohort and phase figures off CAROL's public query API; fetch-avall.mjs reads the probable-cause coding from the NTSB bulk database (avall.mdb, the Events_Sequence defining occurrence); fetch-faa.mjs parses the FAA activity survey for general-aviation and Part 135 flight hours; and a committed file (data/part121_activity.json) carries the NTSB air-carrier hours - hand-downloaded, because the NTSB and BTS hosts block automated fetches, but real and cited. The swap-points are documented at src/lib/source.ts, and the full recipe in HANDOFF.md.
Generated 2026-07-11 15:46 UTC
Source: NTSB Aviation Investigation Search (CAROL) + bulk Aviation database (avall); FAA GA & Part 135 Activity Survey + NTSB air-carrier statistics for flight-hours exposure