TRACE THE
FAILURE
AVIATION FAILURE

The speeds came back.
The plane didn't.

On June 1, 2009, ice crystals likely blocked Air France 447's airspeed probes over the Atlantic. The readings were valid again within about a minute. By then the A330 had climbed out of its flight envelope into a stall, and the warning built for that danger went quiet when the angle was most extreme.

Follow the evidence
Illustrative AI image of a plain white twin-engine airliner, nose raised, falling through night cloud above the Atlantic; not a photograph of Air France 447.
Illustrative AI image from the episode; not a photograph or recording of Air France Flight 447.
THE COMPANION FILMAir France 447: How a Working Airliner Fell From 38,000 Feet Watch episode 6:42

Why did Air France 447 crash?

Air France Flight 447, an Airbus A330 flying from Rio de Janeiro to Paris, fell into the Atlantic on June 1, 2009. All 228 people aboard died. BEA final report, synopsis, p. 17.

The trouble began when ice crystals likely blocked the airspeed probes in cruise and the speed readings failed. That fault was temporary: within about a minute, all three speed displays agreed again. In that same minute, the pilot flying had pulled the airplane into a climb that took it out of its flight envelope. It stalled, and it fell for more than three minutes with its nose above the horizon. The French investigation agency, the BEA, concluded that the crew never understood the airplane was stalling and so never applied a recovery. BEA findings and causes, pp. 198–200.

Watch the Air France 447 investigation

The episode rebuilds the climb, the stall and the stall warning from the BEA’s flight-recorder data, with simplified 3D reconstructions and illustrative scenes.

Watch “Air France 447: How a Working Airliner Fell From 38,000 Feet” on YouTube.

When the speeds vanished

The flight was cruising at 35,000 feet. Its route crossed a cluster of storm cells in the intertropical convergence zone, a situation the BEA found was not exceptional for June. The captain had left the cockpit to rest at about 02:02 UTC. Two copilots were at the controls, and the one in the right seat was flying. BEA §1.1, p. 21; findings, p. 197.

At 02:10:05 the autopilot, then the autothrust, disconnected. The speed shown on the left screen fell sharply from about 275 knots to about 60. The flight controls changed from normal law to alternate law, a backup mode in which the airplane no longer protected itself from reaching too high an angle of attack. BEA §1.1, p. 22; §1.16.3.3, p. 90. The BEA described the obstruction of Pitot probes by ice crystals in cruise as a phenomenon “known but misunderstood by the aviation community at the time.” BEA §3.2, p. 199.

The pilot flying made a nose-up input that raised the pitch to 11 degrees in ten seconds, and the airplane began to climb. The other copilot asked several times for a descent. The pilot flying made nose-down inputs that reduced the pitch and the rate of climb, but the airplane kept climbing. Both copilots noticed and called out the lost speeds; neither called the procedure for unreliable airspeed. BEA §1.1, p. 22; findings, p. 198; §2.2.2, p. 184.

The speed on the left screen was wrong for 29 seconds and valid again at about 02:10:36, showing 223 knots. About 15 seconds after 02:10:51, all three displayed airspeeds were consistent. Since the climb began, the airplane had lost about 50 knots. BEA findings, p. 198; §1.1, pp. 22–23.

Fifty-four seconds of warning

At 02:10:51 the stall warning began to sound continuously, with the airplane at about 37,500 feet and still climbing. A stall has nothing to do with the engines, which worked normally throughout. It happens when the wing meets the air at too steep an angle, the angle of attack, and stops producing enough lift. Buffet began about two seconds into the warning, and the pilot flying selected takeoff/go-around thrust while continuing to make nose-up inputs. BEA §1.1, p. 22; §1.16.4.2, pp. 92–93; findings, pp. 198–199.

The BEA found that nine seconds into the warning the airplane had climbed above its propulsion ceiling, and from then on only a nose-down input could have brought it back within its flight envelope. It topped out at 37,924 feet at 02:11:10, with a pitch attitude and angle of attack of about 16 degrees, and began to fall. Even at full thrust, the drag at that angle of attack kept it from accelerating. BEA §2.1.3.2, p. 179; §1.1, p. 23.

The captain re-entered the cockpit at about 02:11:42, about a minute and a half after the autopilot disconnected. Seconds later the stall warning stopped. It had sounded for 54 seconds. The airplane was near 35,000 feet, falling at about 10,000 feet a minute, with an angle of attack above 40 degrees. BEA §1.1, p. 23; findings, p. 198.

Why did the stall warning go quiet?

The stall warning trusted its angle-of-attack sensors only when the measured airspeed was high enough for a valid reading. If the airspeed from all three air data computers fell below 60 knots, the system treated the angle values as invalid, and the warning stopped. The rule was there to prevent false alarms. BEA §1.6.11, p. 44; findings, p. 198.

In the stall, the measured speeds dropped below that line. Recovery meant lowering the nose; when the nose came down, the angle readings became valid again and the warning came back. The BEA wrote that “it appears that the PF reacted, on at least two occasions, with a nose-up input,” which stopped the warning again. The pilots had no direct display of the angle of attack. BEA §2.3.3, p. 190; findings, p. 198.

The investigators did not present that on-and-off warning as the cause. They judged that by the time it was switching on and off, it was “doubtless already too late” to recover, and no valid angle-of-attack value was below 35 degrees until the end of the flight. The recordings ended at 02:14:28. The airplane hit the water intact, with its nose 16.2 degrees above the horizon, descending at about 10,900 feet per minute. BEA §2.3.3, p. 190; §1.1, p. 24; §1.12.4, p. 81.

The wreckage was found on April 2, 2011, about 3,900 metres down. The flight data recorder was raised on May 1 and the cockpit voice recorder the next day, almost two years after the accident. BEA §1.1, p. 24; p. 20.

Why did no one call it a stall?

The BEA gave several possible reasons the crew did not take the stall warning into account. BEA §3.2, p. 201:

  • The copilots’ stall practice in the simulator had been at low altitude, about 10,000 feet. BEA §2.2.4, p. 185
  • Two brief warnings right after the autopilot disconnected could have seemed false, and no specific display confirmed an approach to stall.
  • Their training and some documents linked buffet to overspeed as well as to the stall. On the A330, buffet occurs only near the stall.
  • The flight director bars, which disappeared and reappeared several times, may have led them to believe their inputs were appropriate.
  • More broadly, the copilots had had no high-altitude training in flying the airplane by hand.

Beyond the cockpit, the operator, the manufacturer and the authorities considered the risk of iced probes mitigated because a procedure for unreliable airspeed existed. As a result, the BEA found, the fact that earlier crews had repeatedly failed to apply that procedure was not considered or analysed before the accident. BEA §2.4, p. 190; §3.2, p. 200.

What changed after Air France 447

On August 31, 2009, three months after the crash, the European Aviation Safety Agency ordered A330 and A340 operators to replace the model of airspeed probe that Flight 447 carried. EASA AD 2009-0195.

The BEA addressed 41 safety recommendations. They included specific and regular training in manual handling of the approach to stall and stall recovery, including at high altitude, and an evaluation of an angle-of-attack indicator directly accessible to pilots. BEA synopsis, p. 17; §4.2, pp. 204–205. Its recommendation to extend recorder locator-beacon transmission to 90 days is now in European rules: Regulation (EU) 2015/2338 requires flight recorders on commercial air transport aeroplanes to carry underwater locating devices that last at least 90 days, from June 16, 2018. BEA §4.1.1, p. 203; Regulation (EU) 2015/2338.

On May 21, 2026, the Paris Court of Appeal found Air France and Airbus guilty of involuntary manslaughter, overturning their 2023 acquittal, and fined each company €225,000, the maximum for a corporation. Both companies have appealed to the Cour de cassation, France’s highest court, which reviews points of law rather than the facts. As of October 16, 2026, that appeal is pending. Reuters via BusinessDay; Airbus statement.

We remember the 228 people who died aboard Air France Flight 447, and everyone who loved them.

For another investigation into how a sensor problem and an automated system’s logic confused a crew, read what went wrong on Lion Air Flight 610.

Primary sources and visual provenance

  1. BEA (Bureau d’Enquêtes et d’Analyses), final report on the accident on June 1, 2009, to the Airbus A330-203 F-GZCP, flight AF 447, English edition (July 2012): synopsis, p. 17; history of the flight, pp. 21–24; stall warning logic, §1.6.11, p. 44; flight control laws and warnings, pp. 89–93; analysis, pp. 179–190; findings and causes, pp. 197–201; recommendations, pp. 203–205. Page numbers are the report’s printed pages. Report PDF.
  2. EASA Airworthiness Directive 2009-0195, Airbus A330 and A340 Pitot probes, replacement (August 31, 2009). AD PDF.
  3. Commission Regulation (EU) 2015/2338, flight recorders, underwater locating devices and aircraft tracking. Regulation PDF.
  4. Reuters via BusinessDay (May 21, 2026) and the Airbus press release (May 2026) on the appeal-court verdict and the appeal to the Cour de cassation.

The article image and preview thumbnail are illustrative AI images from the episode; the thumbnail adds editorial typography. Neither is a photograph or recording of Flight 447. The episode uses AI narration, AI-generated illustrative scenes and simplified 3D reconstructions built from the BEA’s recorded flight data; the BEA and EASA pages and BEA photographs shown in it are real.