The accident record, counted
Independent and not affiliated with Cirrus Aircraft — not a Cirrus Training Center, and no training is offered here.
The safety page argues that the Cirrus accident record improved, and that what changed was training culture rather than the aeroplane. That is a claim, and claims should be checkable. This page is the arithmetic behind it, taken from the NTSB's public accident database — 558 accidents between 2001 and 2025, of which 186 were fatal. Counts only: no individual accident, aircraft or person appears anywhere on this page.
The rate, and the denominator problem
Accident rates are normally quoted per 100,000 flight hours, and the safety page declines to quote one for a good reason: nobody measures fleet hours, they estimate them, so two honest analyses of the same period can disagree. That objection stands and this page does not pretend to have solved it.
What it does instead is use a denominator that is exactly countable. Every Cirrus that has ever held a US registration can be given a span — from the year it was built to the year its registration was cancelled, or to now if it is still flying — and adding those spans up gives the number of aircraft on the register in each year. Accidents per 1,000 registered aircraft-years is then a real division of two counted numbers, with no estimation anywhere in it.
| Period | Aircraft-years | Accidents | per 1,000 | Fatal | per 1,000 |
|---|---|---|---|---|---|
| 2001–2007 | 11,981 | 81 | 6.76 | 32 | 2.67 |
| 2008–2014 | 28,361 | 162 | 5.71 | 69 | 2.43 |
| 2015–2021 | 38,694 | 171 | 4.42 | 51 | 1.32 |
| 2022–2025 | 31,778 | 144 | 4.53 | 34 | 1.07 |
Read the last two columns against each other, because that is where the story is. The overall accident rate fell by about a third. The fatal accident rate fell by 60%. Cirrus pilots did not stop having accidents — they are still bending aeroplanes at a broadly similar rate. What changed is that the accidents stopped killing them. That is precisely the signature you would expect from a parachute that gets used and from training that teaches when to use it, and it is not the signature of better handling or a safer airframe, which would have shown up as fewer accidents rather than more survivable ones.
The improvement was a step, not a slope
The shape of that line is worth more than its endpoints. An aeroplane getting gradually better, or a fleet slowly ageing into experience, produces a gentle downward drift. This is not that. The rate moves around inside a band for a decade without trending, drops hard across 2013–2015, and then holds flat near 1.1 for another decade. Something changed in how these aeroplanes were being flown, over about three years, and then the new level persisted. The page that explains what is the safety page; this one establishes that it happened.
What the accidents actually were
The NTSB codes a defining event for each accident. Of the 477 accidents recorded under the current taxonomy, 460 have one.
| What the accident was | Accidents | Fatal | Share fatal |
|---|---|---|---|
| Loss of control in flight | 108 | 62 | 57% |
| Loss of engine power (total) | 53 | 4 | 8% |
| Loss of control on ground | 45 | 1 | 2% |
| Loss of engine power (partial) | 22 | 1 | 5% |
| Aerodynamic stall/spin | 19 | 12 | 63% |
| Unknown or undetermined | 19 | 11 | 58% |
| Runway excursion | 17 | 1 | 6% |
| Controlled flight into terrain | 15 | 12 | 80% |
| Abnormal runway contact | 15 | 2 | 13% |
| Powerplant system malfunction | 12 | 0 | 0% |
| Hard landing | 12 | 1 | 8% |
| VFR encounter with IMC | 10 | 10 | 100% |
Three rows carry most of the meaning.
- Loss of control in flight is the killer — 108 accidents, 62 of them fatal, a 57% fatality rate and far more than any other common category. Add aerodynamic stall/spin at 63% fatal and controlled flight into terrain at 80%, and the pattern is the ordinary killers of general aviation, exactly as the safety page says — not structural failure, not the parachute, not the avionics.
- Engine failure has largely stopped being fatal. A total loss of engine power is the second most common accident at 53 events, and only 4 of them killed anyone — 8%. In a conventional single, an engine failure without a landing site is a classic fatal scenario. This is the single clearest number on the page for what CAPS changes, and it is why the CAPS page is about the decision rather than the hardware.
- VFR into IMC is 10 for 10. Ten accidents, ten of them fatal, no survivors. It is not the most common way to crash a Cirrus and it is comfortably the most lethal. That is the entire argument for the instrument rating on the ratings page, in one line of a table.
Weather
Instrument conditions account for 74 of the accidents but 52 of the fatal ones — about 13% of all accidents and 28% of the ones that killed people. Put the other way round: an accident in IMC was fatal 70% of the time, against 24% in visual conditions. Weather does not cause many more accidents in this type; it changes how the ones it causes end.
Training flights
Accidents on instructional flights were 70 of the total, of which 15 were fatal — 21%, against 33% for personal flying. Training flights crash, and they crash in the circuit doing circuit things; they kill people considerably less often. Be careful what you read into that — an instructional flight is in good weather, near an airport, with two pilots aboard, which is a different exposure rather than purely a different standard of flying.
What this page does not publish
No individual accident appears here: no registration, no date, no location, no narrative, no probable cause. Every row behind these counts is a real accident and most of the fatal ones have a named pilot in the NTSB narrative, so the script that produced this page reads those fields to identify the aircraft type and then discards them. The same reasoning keeps photographs off the CAPS page: a browsable list of Cirrus crashes would be the written version of a wreckage photograph.
If you want a specific event, the NTSB publishes every one of them with full narratives and probable-cause findings at CAROL, which is the authoritative source and is free to search.
Method, and four limits
Source: the NTSB's bulk aviation accident database, files published 1 September 2026, combining the current file with the pre-2008 archive. Records were matched on manufacturer and model — both free text, so the match normalises spellings and deliberately excludes the Schempp-Hirth "Standard Cirrus" glider and the VK30 homebuilt, which share the name. Incidents are excluded; only accidents are counted. The denominator comes from the FAA register exactly as on the fleet-data page, using year of manufacture rather than certificate date, because a registration certificate is reissued on every change of ownership and would date a 2004 airframe sold in 2023 to 2023.
- These rates are not comparable to published ones. Per aircraft-year is not per flight hour. Nor is the figure safely comparable between types: a flight-school Cessna 172 flies several times the hours of a privately owned SR22, so per-aircraft-year would quietly flatter the Cirrus. It is used here for the trend within one type over time, and for nothing else.
- The cause breakdown is 2008 onward only. The NTSB replaced its occurrence taxonomy around 2008 and the older coding is not the same vocabulary, so merging the two would manufacture a trend. The rate table above does span the whole period, because counting accidents and deaths does not depend on the taxonomy.
- Recent years are provisional. Investigations stay open for a long time, so the most recent years can gain accidents and change classification after the fact. The 2025 accident count in particular is high enough to be worth treating as unsettled; the fatal counts, which drive the trend, are the more stable figure.
- US data only. The NTSB investigates US accidents and accidents involving US-registered aircraft abroad. A Cirrus on a European register that crashes in Europe is not in any of this.