Inadvertent Go-Around Mode Activation Leads to Fatal Nose-Dive
The Boeing 767 was descending normally toward Houston when, suddenly, its automated flight mode changed.
The aircraft was about 40 miles southeast of George Bush Intercontinental Airport, descending through roughly 6,300 feet. The weather ahead included clouds and turbulence, but nothing suggested that the airplane was seconds away from accident.
Then the go-around mode activated.
The thrust levers advanced.
The nose began pitching upward.
The first officer, who was flying the aircraft, did not recognize what had happened.
Instead, he reacted as though the airplane were stalling.
He pushed the control column forward.
But rather than recognizing the acceleration as evidence that the airplane was diving, the first officer became convinced that he was dealing with a stall. He continued applying nose-down elevator.
The 767 rapidly descended.
The captain finally realized something was seriously wrong and attempted to intervene.
It was too late.
Only 32 seconds separated the inadvertent activation of go-around mode from the aircraft's impact with Trinity Bay. All three people aboard Atlas Air Flight 3591 were killed. (NTSB)
The final cockpit recording captured the first officer asking where his airspeed had gone, followed by his belief that the airplane was stalling. The captain then desperately attempted to recover the aircraft as ground-proximity warnings sounded. Contemporary reporting later described the transcript as revealing the crew's increasingly confused and desperate final moments. (KPRC)
It was in a steep, accelerating dive.
And investigators eventually determined that an inadvertent activation of the Boeing 767's go-around mode, combined with the first officer's response and the captain's failure to take effective control, created the fatal chain.
A Routine Cargo Flight to Houston
Atlas Air Flight 3591 was a scheduled domestic cargo operation from Miami International Airport to George Bush Intercontinental Airport in Houston.
The aircraft was a Boeing 767-375BCF, registration N1217A, operating for Atlas Air on behalf of Amazon.com Services LLC and carrying cargo for Amazon and the U.S. Postal Service. (NTSB Data)
First Officer Conrad Jules Aska, 44
Captain Sean Archuleta, 36, a non-revenue pilot occupying the jump seat
All three died in the crash. Contemporary Houston reporting identified the men and described Archuletas a Mesa Airlines captain who was traveling toward Houston before beginning a new job with United Airlines. (Houston Chronicle)
The aircraft departed Miami at approximately 10:33 a.m. Central Standard Time.
For most of the flight, everything was normal.
The 767 was descending toward Houston under instrument flight rules, and Houston Center was providing radar vectors toward the airport.
The accident appeared, at first, to be an ordinary approach through changing weather.
Then the crew encountered turbulence.
The Approach Begins Normally
The aircraft began its descent toward Houston from cruising altitude.
According to the NTSB's investigation, the airplane's flight path remained normal for most of the descent. The aircraft was configured appropriately, maintaining normal speed and pitch as it approached the Houston area. (NTSB Data)
The crew was communicating with air traffic control and receiving radar vectors for the approach to Runway 26L.
Ahead of them was a band of weather associated with a cold front.
The conditions included clouds, turbulence and instrument meteorological conditions.
The NTSB later determined that the aircraft was descending through approximately 6,300 feet when the critical event began. (NTSB)
There was no engine failure.
There was no structural breakup.
There was no loss of fuel.
Then, at approximately 12:38:31 p.m., something changed.
A Button Was Activated That Should Not Have Been
Go-around mode is designed to help pilots execute a missed approach.
When activated on the Boeing 767, the system commands significant changes in aircraft automation. The autothrottle increases engine thrust, while the flight guidance system commands a nose-up pitch.
It is exactly what a crew wants when it has decided to abandon a landing and climb away.
But Flight 3591 was not going around.
Neither pilot called for a go-around.
Neither pilot verbally acknowledged that go-around mode had been selected.
And there was no operational reason for the aircraft to suddenly enter that mode.
The NTSB determined that the activation was inadvertent. Investigators concluded that the first officer most likely accidentally contacted one of the go-around switches while manipulating the nearby speed-brake lever, possibly with his wrist or wristwatch. (NTSB)
The result was immediate.
The aircraft's thrust increased.
Its pitch began moving upward.
The automation was doing exactly what it had been designed to do.
The problem was that the pilots did not realize what the airplane was doing.
The First Officer Sees Something He Does Not Understand
The first officer, Conrad Aska, was the pilot flying.
He was responsible for manually controlling the aircraft while the captain served as pilot monitoring.
The go-around system commanded an increase in thrust and a nose-up pitch.
But the acceleration created a dangerous sensory illusion.
This is a form of spatial disorientation in which forward acceleration can make a pilot perceive the aircraft as pitching upward more than it actually is. The sensation can be extremely convincing, particularly when the pilot is flying in clouds and cannot use the outside horizon as a reference. (NTSB)
The first officer appears to have interpreted the aircraft's unexpected pitch and acceleration as evidence of a stall.
His reaction was therefore the opposite of what the aircraft actually needed.
He pushed forward.
The nose went down.
"Where's My Speed?"
The cockpit voice recorder captured the first officer's confusion.
Then he repeated the concern.
The aircraft, however, was not approaching a stall.
The flight-data recorder later showed that airspeed was actually increasing rapidly.
This contradiction became one of the most important pieces of evidence in the investigation.
A stall occurs when the wing exceeds its critical angle of attack and can no longer produce sufficient lift.
Flight 3591 was doing almost the opposite.
Its nose was going down.
Its speed was increasing.
The NTSB's analysis found no evidence that the aircraft's stall-warning system had activated before the first officer began his nose-down inputs. The data were inconsistent with an aerodynamic stall. (NTSB Data)
The airplane was not falling because it had stalled.
It was falling because the flight crew was commanding it into a steep descent.
And the first officer did not realize it.
The Dive Accelerates
Once the first officer pushed forward on the control column, the aircraft began pitching farther downward.
The autopilot had been overridden by the manual elevator inputs.
The nose continued downward.
That acceleration reinforced the first officer's perception that something was wrong with the aircraft's speed and flight condition.
The NTSB found that the first officer continued making nose-down inputs even as the aircraft entered an increasingly steep descent. (NTSB)
This created a severe feedback loop.
The airplane pitched up because go-around mode had activated.
He pushed the nose down.
The acceleration produced a sensory illusion.
He pushed farther.
The airplane dived more steeply.
And the situation became increasingly difficult for the captain to understand and correct.
The entire sequence lasted only seconds.
The Captain Was Slow to Recognize the Emergency
The captain, Ricky Blakely, was the pilot monitoring.
His responsibility included monitoring the airplane's flight path, instruments, automation and the actions of the pilot flying.
But the NTSB determined that he did not adequately recognize the developing loss of control or take positive control of the airplane quickly enough to prevent the crash. (NTSB)
This was a crucial part of the final accident sequence.
The captain did not immediately understand why.
The two pilots were therefore applying opposing inputs.
The captain eventually pulled back.
The Boeing 767's elevator control system allowed opposing control inputs to produce different elevator positions on either side of the aircraft. The NTSB found that the opposing inputs continued for approximately 10 seconds as the dive became increasingly severe. (Skybrary)
During those critical seconds, the airplane was losing altitude at an extraordinary rate.
The crew had only moments left.
The Airplane Was Now in a Steep Dive
The 767's descent became almost vertical compared with its normal flight path.
The airplane's pitch eventually reached approximately 49 degrees nose-down.
The airspeed increased dramatically.
The aircraft descended through the clouds toward Trinity Bay.
A security camera near Anahuac captured the final portion of the flight. The video showed the aircraft emerging from the weather in an extreme nose-down attitude before disappearing behind the terrain and water. Contemporary reporting described the footage as showing a steep dive. (KPRC)
Witnesses on the ground initially had no explanation for what they had seen.
The airplane appeared to be falling almost vertically.
There was no obvious engine fire.
No visible explosion preceded the impact.
The Boeing 767 simply descended toward the bay at very high speed.
"Captain!"
Ground-proximity warning systems issued warnings.
The captain called for the airplane to pull up.
But by this point, the aircraft was too low.
That number is perhaps the most important fact in the entire accident.
Thirty-two seconds.
That was all the crew had.
The aircraft had gone from a normal descent to an unrecoverable dive in roughly half a minute.
At approximately 12:39 p.m., Flight 3591 crashed into the shallow, muddy waters of Trinity Bay, approximately 41 miles southeast of Houston's George Bush Intercontinental Airport. (NTSB Data)
There were no survivors.
The Impact in Trinity Bay
The crash site was in shallow marshland and bay water near Anahuac, Texas.
The force of the impact fragmented the Boeing 767 and scattered wreckage over a large area.
The recovery effort was complicated by the muddy terrain and water.
Local authorities initially recovered two victims while continuing the search for the third. Captain Ricky Blakely's remains were recovered several days later. Contemporary Houston Chronicle reporting documented the recovery operation and identified all three victims. (Houston Chronicle)
The crash also left a profound impact on the local community.
Chambers County Sheriff Brian Hawthorne described the sight of a Boeing 767 crashing near Anahuac as something he had never experienced during his decades in law enforcement. The aircraft had fortunately come down in the bay rather than the nearby populated area. (FOX 26 Houston)
That geographic fact prevented additional casualties on the ground.
But for the families of the three men aboard, the loss was severe.
The Victims
Captain Ricky Blakely was an experienced airline pilot who had accumulated more than 11,000 flight hours, including substantial experience on the Boeing 767. He had joined Atlas Air in 2015 and became a 767 captain in 2018. (Wikipedia)
First Officer Conrad Aska had more than 5,000 total flight hours and approximately 520 hours on the 767. He had previously flown several aircraft types and had worked for multiple airlines. (Wikipedia)
Jumpseat pilot Sean Archuleta was a Mesa Airlines captain who was traveling toward Houston. Contemporary reporting noted that he was a new father and was preparing to begin a new career chapter with United Airlines. (CBS News)
The accident therefore was not the loss of an inexperienced crew operating an unfamiliar aircraft.
It involved experienced professional aviators.
That made the human-factors findings especially important.
The Investigation Begins
In the days following the crash, investigators initially had very little information.
The wreckage was spread across the bay.
The aircraft had descended so rapidly that there was no traditional distress call explaining the problem.
Then investigators recovered the cockpit voice recorder.
The NTSB announced on March 1 that the CVR had been recovered from the crash site and transported to its laboratories in Washington, D.C. (KPRC)
The recorder contained approximately two hours of cockpit audio.
The quality was imperfect, but investigators were able to reconstruct the crew's final communications.
The flight data recorder provided the other half of the story.
Together, the two recorders showed that the aircraft had not simply entered an unexplained stall.
Something had happened to the aircraft's automation.
And then something had happened inside the cockpit.
The Go-Around Switch Becomes the Key
The flight-data recorder showed that the go-around mode had activated at approximately 12:38:31 p.m.
The airplane was about 6,300 feet above mean sea level.
It was still descending.
There was no reason for a go-around.
Neither pilot announced one.
The activation was therefore inconsistent with the phase of flight and cockpit communications. (NTSB)
Investigators examined the physical location of the go-around switches.
On the Boeing 767, the switches are located near the thrust levers.
The speed-brake lever is nearby.
It was not an intentional go-around.
It was a mode change that neither pilot immediately recognized.
And because the aircraft's automation responded normally to the command, the cockpit suddenly presented the pilots with an unexpected combination of thrust and pitch.
The Illusion That Changed Everything
The human brain normally combines visual information, vestibular sensations and instrument indications to determine aircraft attitude.
In clouds, visual information is greatly reduced.
That makes the vestibular system more important—but it is not reliable during acceleration.
When an aircraft accelerates forward, the inner ear can interpret the acceleration as a pitch-up sensation.
The critical point is that the airplane was actually accelerating because the engines had advanced.
But the first officer's sensory perception suggested that the airplane's attitude was wrong.
He responded by pushing forward.
This is why the accident is such a powerful human-factors case study.
The first officer was not deliberately diving the airplane.
He was attempting to correct what he believed was a dangerous flight condition.
His perception was wrong.
And once he began responding to that false perception, the aircraft's actual trajectory made the illusion even harder to escape.
Training Problems Were Already Known
The NTSB investigation went beyond the accident flight and examined the pilots' training histories.
Investigators found that First Officer Aska had experienced significant difficulties during previous training.
He had failed his first attempt at Boeing 767 qualification and required additional training before successfully completing the process. Contemporary reporting based on the NTSB docket described concerns about his ability to anticipate problems and maintain situational awareness under stress. (Houston Chronicle)
More significantly, investigators discovered that Aska had previously failed to complete training at two other airlines and had not disclosed portions of that employment history to Atlas Air.
That meant Atlas Air did not have access to all of the information that could have been relevant when evaluating his suitability for employment.
The NTSB eventually identified this as part of a broader systemic problem rather than simply an individual failure. (NTSB)
The accident therefore raised questions about how airlines select pilots, evaluate training performance and identify pilots who may struggle under high workload or unexpected conditions.
The Captain's Role
The investigation also examined Captain Blakely's actions.
The NTSB did not conclude that the captain caused the initial event.
But the captain was responsible for monitoring the aircraft and intervening when the flight became unsafe.
This is a fundamental Crew Resource Management lesson.
The pilot monitoring cannot simply watch the instruments.
He must continuously assess whether the aircraft's behavior makes sense.
If the airplane suddenly pitches up, accelerates and departs from its expected flight path, the monitoring pilot must identify the discrepancy.
If the pilot flying is making inappropriate control inputs, the monitoring pilot must intervene decisively.
In Flight 3591, those safeguards did not work quickly enough.
The NTSB's Probable Cause
After an extensive investigation, the National Transportation Safety Board issued its final findings.
The official probable-cause statement reads:
— National Transportation Safety Board, DCA19MA086 / NTSB Aircraft Accident Report AAR-20/02. (NTSB Data)
The NTSB identified several contributing factors.
The Board also identified systemic deficiencies in the aviation industry's pilot selection and performance-measurement practices, which failed to address the first officer's aptitude-related deficiencies and maladaptive stress response.
Finally, the FAA's failure to implement the Pilot Records Database in a sufficiently robust and timely manner was also identified as a contributing factor. (NTSB)
The finding was therefore much broader than "pilot error."
The accident exposed weaknesses at several levels of the aviation system.
Accident Facts
- Flight
- Atlas Air 3591
- Aircraft
- Boeing 767-375BCF
- Registration
- N1217A
- Operator
- Atlas Air
- Operation
- Cargo flight
- Date
- February 23, 2019
- Departure
- Miami International Airport
- Destination
- George Bush Intercontinental Airport, Houston
- Accident location
- Trinity Bay, Texas
- Accident time
- Approximately 12:39 p.m. CST
- Altitude when go-around activated
- Approximately 6,300 ft MSL
- Occupants
- 3
- Fatalities
- 3
- Survivors
- 0
- Defining event
- Inadvertent go-around mode activation followed by loss of control
- Aircraft damage
- Destroyed
- NTSB investigation
- DCA19MA086
- Final report
- NTSB/AAR-20/02
- Time from go-around activation to impact
- Approximately 32 seconds
Aviation Safety Network independently records the February 23, 2019 accident involving Boeing 767-375ER(BCF) N1217A near Trinity Bay, with three fatalities and Amazon Air listed as the operation with Atlas Air. (Flight Safety Foundation)
Why the Final Seconds Are So Important
Flight 3591 is particularly revealing because investigators had multiple independent sources of evidence.
The flight-data recorder showed the airplane's actual attitude, airspeed, altitude and control inputs.
ATC recordings established the aircraft's expected flight sequence and communications.
Radar data showed the rapid descent.
Together, those sources eliminated several possible explanations.
The airplane did not suddenly lose both engines.
It did not experience a conventional aerodynamic stall.
It did not suffer a mysterious structural breakup.
The critical sequence began when the go-around mode was inadvertently activated.
The first officer then reacted incorrectly to the aircraft's changing attitude and acceleration.
The captain failed to intervene effectively.
The Broader Aviation Safety Lesson
Atlas Air Flight 3591 is one of the clearest demonstrations of how automation can create an unexpected situation without actually malfunctioning.
The Boeing 767's go-around system did exactly what it was designed to do.
It commanded nose-up pitch.
It attempted to put the aircraft into a go-around.
The problem was that the mode had been activated at the wrong time, and the pilots did not recognize what had happened.
That created a dangerous mismatch between what the aircraft was doing and what the pilot believed it was doing.
The captain, meanwhile, did not immediately establish positive control.
By the time the crew recognized the true situation, the airplane was already descending at an extreme rate.
The entire sequence took just 32 seconds.
It is not established that automation is dangerous.
The lesson is that pilots must always understand which mode the airplane is in, what the automation is commanding and whether the aircraft's actual behavior matches what they expect.
It is also a lesson in Crew Resource Management. A pilot monitoring must actively monitor—not simply observe. When the aircraft's flight becomes inexplicable, intervention must be decisive.
And there is a larger organizational lesson.
The NTSB found that Flight 3591 was influenced by weaknesses in pilot selection, performance measurement, records transfer and regulatory oversight. (NTSB)
The tragedy was therefore not created by one button alone.
It was created by a chain of human perception, automation, training, monitoring and organizational defenses.
The first officer's final concern about airspeed was tragically mistaken: the airplane was not running out of speed.
It was accelerating into a dive.
The final warning came as the crew finally understood the danger.
But there was no longer enough altitude to recover.
Three men lost their lives over Trinity Bay that afternoon.
For aviation professionals, the enduring lesson is brutally simple: when an aircraft suddenly behaves differently from what you expect, do not fight the airplane based on a guess. Identify the active automation mode, trust the instruments, communicate the problem, and if necessary, take positive control.
Because sometimes the difference between a recoverable surprise and a fatal loss of control is measured not in minutes—but in seconds.
Primary Source / Investigation Record
Investigation reference: DCA19MA086