For decades, a successful stall exercise looked like this: warning, thrust set to the stop, attitude held, a few dozen feet lost, examiner satisfied. That was the right answer in a check. It was also, almost gesture for gesture, what the recorders found in a series of loss-of-control-in-flight accidents. Upset recovery training was not born of a teaching fashion: it was born of the finding that an exercise which had always been graded was training the wrong reflex.
What the stall exercise did not train
Loss of control has long headed the list of fatal accident categories in commercial air transport, alongside runway excursions and collisions with terrain. That ranking coexisted for a long time with a stall exercise judged satisfactory, repeated at every periodic check, passed by everyone. It is that coexistence which eventually came under scrutiny.
The classic exercise stopped at the approach to the stall, at the first indication: aural warning, stick shaker, buffet. The instruction fitted into three words — thrust, attitude, altitude — and the assessment criterion was the height lost. Three flaws were lodged in it, and it took time to name them.
The first is an inversion of priorities. A wing stalls because its angle of attack exceeds a critical value, not because an airspeed indicator displays a number. Holding altitude at the first indication amounts to maintaining that angle of attack, or even to increasing it if the pilot pulls to compensate for the rising drag. On an aeroplane with underwing engines, setting maximum thrust adds a nose-up pitching moment on top of that, which then has to be countered: the action meant to resolve the situation makes it worse.
The second is a truncated envelope. Since the exercise stopped short of the full stall, nobody saw what comes next: one wing dropping before the other, the ineffectiveness of the ailerons, the attitude tipping over while the control column is already at the aft stop. A crew one day faced with a real stall found itself in front of a scene it had never seen, neither in flight nor in the simulator.
The third is predictability. The exercise was announced, in level flight, in smooth air, in an agreed configuration, at a comfortable altitude, in a known order. Nothing of what makes a loss of control difficult — the unexpected, the noise, the doubt about what the aeroplane is doing, fear — was present. What was being trained was a choreography, not a recovery of control.
The old stall exercise taught pilots to hold altitude when a stall began, but this actually makes the stall worse. Real stalls involve unpredictable situations where the plane’s behavior changes dramatically, and pilots need to react differently than they were trained to do.
In 2019, a Boeing 737 MAX 8 experienced a stall at 24,000 feet during a go-around. The crew initially followed the old reflex of holding altitude, which worsened the stall. Only after pushing forward and reducing angle of attack were they able to recover control. The aircraft had lost 5,000 feet by the time recovery was complete.
Upset: a definition of state, not of cause
The reference definitions adopted by the industry describe a state, with simple limits: a pitch attitude of about twenty-five degrees nose up or ten degrees nose down, a bank angle of about forty-five degrees, or an airspeed clearly inappropriate for the conditions of the moment. The last criterion is the most important and the quickest to be forgotten: an aeroplane can be in an upset with the wings level and the nose on the horizon.
That choice has a direct consequence for training. Because the definition says nothing about the cause, it rules out organising the training by scenario. And the causes have no kinship with one another: the wake of a preceding aircraft, ice on a leading edge, an autopilot handing back a correction it could no longer hold, doubtful airspeed information, an asymmetric failure, a somatogravic illusion during a night go-around, or simply a control input on an aeroplane that is poorly understood. Building the training around the recovery rather than around the scenario is the only way to cover a list that cannot be closed.
An “upset” is when an aircraft is flying in an extreme attitude – very nose high, very nose low, or banked too steeply. It doesn’t matter why it happened – what matters is how to recover from that dangerous state.
In 2009, Air France Flight 447 entered an upset when ice crystals blocked the pitot tubes, causing unreliable airspeed readings. The crew reacted incorrectly, pulling back on the controls and stalling the aircraft. The upset lasted for 3 minutes 30 seconds before the aircraft crashed into the Atlantic Ocean. The recovery training would have taught them to push forward immediately to regain lift.
The first word of the title matters as much as the second: prevention precedes recovery, and it is prevention that takes up the largest part of the programme. Recognising a slow speed decay in a holding pattern, knowing what the autopilot is trimming, refusing an approach profile that requires descending fast and slowly at the same time, watching the buffet margin rather than the indicated airspeed at high cruise altitude: all of this avoids ever having to fly a recovery manoeuvre. A successful recovery remains, in the analysis, a failure of prevention.
The recovery: an order of actions, not a catalogue
The recovery template issued jointly by the large aeroplane manufacturers, and later taken up in approved programmes, fits into two columns. Its value lies in the order, not in the list.

Nose high. Recognise and announce, disconnect the automation — an autopilot left engaged goes on trimming against the input —, apply nose-down pitch, as far as the forward stop if the elevator alone is no longer enough, and complete with trim. Reduce thrust on an aeroplane with underwing engines, whose pitching moment works against the manoeuvre. Then roll, towards the nearest horizon, to a definite bank angle: banking redirects the lift and brings the nose down without imposing a negative load factor on the airframe or on the crew. The aeroplane recovers its speed on a descending path, which is precisely the aim.
Nose low. Recognise and announce, disconnect, then reduce the angle of attack before anything else: a nose-low aeroplane may be stalled, and pulling on a stalled aeroplane merely sustains the stall. Only then bring the wings level by the shortest path, and pull out while watching two limits that arrive together: the manoeuvring load factor, of the order of two and a half times the weight in clean configuration for a transport aeroplane, and the maximum operating speed, which a nose-low attitude reaches within seconds if thrust is left set.
When an aircraft is nose high, pilots must push forward to reduce angle of attack, then roll wings level. When nose low, they must first reduce angle of attack before pulling up. The key is doing these steps in the right order.
During a 2018 training flight in an Airbus A320, a pilot experienced a nose-high upset at 30,000 feet. Following the recovery procedure, they pushed forward to 15° nose down, rolled to 30° bank, and reduced thrust. The aircraft recovered at 25,000 feet with no damage. The key was following the correct sequence of actions.
Two actions come up regularly in debriefing. The first is the immediate pull-out at a steep bank angle: pulling while the lift is pointing sideways does not straighten the flight path, it tightens the spiral and adds load factor without changing the descent angle at all. The second is the use of the rudder as a roll control. On a transport aeroplane it is not one: it serves symmetry, notably with an engine failed. Alternating, full-travel yaw inputs generate loads on the fin that the certification envelope does not guarantee will be withstood, and that limit is not intuitive from the flight deck.
| Assessment point | Historic approach-to-stall exercise | Revised recovery manoeuvre |
|---|---|---|
| Trigger | First indication: warning, shaker, buffet | The same indication, but the exercise may continue to the full stall where the device is qualified for that envelope |
| First action | Maximum thrust, attitude held | Reduction of the angle of attack, nose-down pitch applied until the buffet stops |
| Thrust | Set to the stop straight away | Adjusted after the angle of attack, reduced when the engine pitching moment works against the recovery |
| Height lost | Pass criterion: minimum loss | An accepted and managed consequence; it is no longer an assessment criterion |
| Role of roll | Wings held level with the ailerons | A recovery means in its own right when nose high; wings levelled first before any nose-low pull-out |
| Rudder | Used to counter roll | Reserved for symmetry; alternating inputs are a structural matter |
| Predictability | Announced manoeuvre, in level flight, at a comfortable altitude | Unannounced scenario, in varied configurations and phases of flight |
The trap is the trim. A nose-high recovery that follows a long autopilot correction begins with a horizontal stabiliser trimmed nose up. The elevator alone may then produce nothing: the crew pushes, feels nothing happening, and concludes that the manoeuvre does not work, where it should conclude that an action is missing. Training to apply nose-down pitch without ever looking at the trim position is training half a manoeuvre — and the other half is not discovered in flight.
What the simulator can do, and what it cannot
A simulator reproduces an aerodynamic model validated against flight test measurements. Within the measured envelope, the reproduction is faithful and the exercise has training value. Beyond it, the model extrapolates. As long as qualification stopped at the first stall indication, an exercise pushed further produced behaviour that was plausible but not guaranteed: the crew learned there a recovery that might not exist, which is soberly called negative training.
The response was twofold. On the one hand the extension of the qualification criteria, to cover the envelope beyond the stall on the basis of testing and wind tunnel work. On the other the instructor’s tooling, which now includes an indication when the device leaves its validated envelope — information that only means something if the instructor knows what to do with it, and that is a point to be checked in the standardisation programme.
What remains is what no extension of the model will settle. A motion platform reproduces the onset of an acceleration, then quietly returns to its neutral position: it cannot hold a load factor for several seconds, nor produce a load factor below one. A real nose-low recovery, with a load factor of the order of twice the weight sustained for several seconds, is felt in the simulator as a non-event. And the trainee knows, at every moment, that the machine cannot kill him.
Startle, surprise, and the part only the aeroplane teaches
Two distinct phenomena are often confused. Startle is a physiological reflex triggered by an abrupt stimulus: it is counted in fractions of a second, it contracts the body, it briefly cuts off mental availability, and it cannot be trained — it is anticipated and endured. Surprise is cognitive: it is the gap between what one believed about the situation and what it actually is. Its resolution is counted not in fractions of a second but in seconds, sometimes in tens of seconds, and it is what consumes the time available.
It is for that portion that advanced training is carried out in flight, on an aerobatic-capable aeroplane rather than in a simulator. Not to teach aerobatics, but to install a physiological reference: having already felt a sustained load factor, having already experienced the disorientation that follows a prolonged rotation, having already gone on reading instruments with the horizon askew. When the day comes, the sensation will not be a first, and that is all that is asked of it.
What transfers is therefore not the manoeuvre. A light aeroplane completes a roll in a fraction of the time a long-haul aircraft needs, with forces and a delicacy of touch that bear no comparison; nothing that belongs to the fineness of the input carries over from one aeroplane to the other. What transfers is the human part: tolerance to the sensation, the ability to keep reading, and an order of actions installed firmly enough to survive a moment when reasoning is no longer available.
The right habit: state the situation before correcting it. Calling out aloud what the artificial horizon shows — nose high, banked right, speed decaying — costs a few seconds and produces three effects: a shared diagnosis, a ban on flying the sensation rather than the instrument, and an entry point for the pilot monitoring to call out the trim, the thrust and the height. The crews that recover cleanly are not those who act fastest, but those who name fastest.
Where this training fits into a career path
For a young pilot, the requirement does not appear all at once: it is spread across the whole progression. The theoretical knowledge is covered from the common core onwards. Basic elements are built into practical flight training during the professional course. Advanced training, for its part, is a course in its own right, delivered in an aeroplane by an organisation approved for it and by an instructor holding the corresponding privilege: of the order of five hours of theoretical instruction and three hours of dual flight under the approved programme, sanctioned by a certificate of completion.
That certificate is a condition of entry to a multi-pilot type rating: without it, enrolment is refused. It therefore comes before the first transport aeroplane, at a point when the pilot has no habits yet to undo. Next comes the type-specific part, covered in the simulator during the rating, then the repetition under the operator’s recurrent training and checking programme, whose multi-year cycle must cover these elements. Organisations have neither the same training fleet nor the same depth of programme, and it is worth comparing the training paths on offer before committing.
| Component | Ground instruction | Flight simulator | Aerobatic-capable training aeroplane |
|---|---|---|---|
| Sustained load factor | No | No: transient reproduction, then washout | Yes, within the limits of the aeroplane |
| Vestibular disorientation | Described | Partial and short-lived | Yes, including its after-effects |
| Surprise effect | None | Real if the scenario is not announced | Real, and redoubled by the stakes |
| Behaviour beyond the stall | Explained | Only where the device is qualified for that envelope | Yes |
| Specifics of the type operated | Yes | Yes, that is its whole purpose | No |
| Automation, protections, trim | Yes | Yes | Not applicable |
What remains to be checked on your own side
Three checks are worth making in your own documentation rather than in an article. The first concerns the assessment criterion: does the check form used by the operator still list height lost as a grading item? It happens, and an old form is enough to reinstall the very reflex the overhaul set out to remove. The second concerns the training device: does its validated envelope extend beyond the stall, or does the exercise deliberately stop earlier? The answer is in the qualification file, not in the programme. The third concerns the type-specific recovery: the generic template is generic, and the aircraft manual may impose a control or configuration peculiarity that cannot be invented in the moment.
A grey area remains, and it is better to name it. The advanced training certificate is issued once, with no expiry, whereas the benefit it attests — a physiological tolerance, a lived reference — erodes over the years. The paper does not expire; the effect does. Some operators fill the gap with unannounced recurrent sessions, others do not, and the question comes up regularly on the professional forums, with no common answer. For anyone preparing for a rating, that changes one concrete thing: the question to put to the organisation is not whether it issues the certificate, but which aeroplane it uses, what envelope it actually explores in flight, and what the trainee will have felt on stepping off the aircraft.
This text describes a mechanism and a training rationale; it replaces nothing. The applicable regulation, the specifications of the training device used, the aircraft flight manual and the operator’s approved documentation prevail over everything above, including where the article appears more explicit than they are.
Questions fréquentes
Is advanced upset prevention and recovery training mandatory for everyone?
In the European system, it is a condition of entry to a first multi-pilot aeroplane type rating: a candidate who does not hold the certificate cannot begin the course. Pilots who already held such a rating before the requirement came into force are in principle not subject to it retroactively, but that point must be checked in the applicable text and in the position of the authority concerned. The subject matter is, however, covered for everyone by the operator’s recurrent training and checking programme.
Can this training be done entirely in a simulator?
Not for the advanced part: it includes an in-flight element precisely to deliver what a simulator cannot produce, namely a sustained load factor, real vestibular disorientation and the physiological after-effects of an upset actually experienced. The simulator remains indispensable for the type-specific part, the automation, the protections and behaviour beyond the stall where its envelope is qualified for it. The two media are complementary, not interchangeable.
Does the certificate issued have an expiry date?
It is a certificate of course completion, issued once, not a rating to be renewed at fixed intervals. Recency comes instead from the operator’s recurrent programme, which covers the subject matter over its cycle. It is an acknowledged grey area: the document does not expire, whereas the physiological reference acquired in flight erodes over the years.
Why is height loss no longer a pass criterion?
Because it made the wrong action rational. Reducing the angle of attack always costs height; grading the height lost therefore amounted to penalising the first correct action and rewarding the one that maintains the angle of attack. Height loss is now treated as a consequence to be managed, and terrain clearance is dealt with in prevention, not during the recovery.
What is the role of the rudder in an upset recovery?
It serves symmetry, notably in the case of asymmetric thrust, and it is not a roll control on a transport aeroplane. Alternating, large-amplitude yaw inputs create loads on the fin that the certification envelope does not guarantee for that kind of reversed loading. Levelling the wings is done with ailerons and spoilers, and the rudder stays with its own function.
Does what is learned on an aerobatic aeroplane really transfer to a transport aircraft?
The manoeuvre, no: roll rates, control forces and delicacy of input have nothing in common between a light aeroplane and a long-haul aircraft. What transfers is human: tolerance to the sensation, the ability to keep reading the instruments in an unusual attitude, and an order of actions installed firmly enough to withstand startle. The type-specific part is then covered in the simulator during the rating.

Leave a Reply
You must be logged in to post a comment.