On the 6th of September 2026, 21 Air flight 7598 landing at Miami International Airport overran the runway, killing five people.

The aircraft was a 32-year-old Boeing 767-300 wide-body airliner converted for cargo, registered in the US as N1997A.

That day, the aircraft had flown Cincinnati-Miami-San Juan with two flight crew aboard. I have not seen any information as to whether it was the same crew.

At San Juan, the aircraft was loaded with about 32,000 pounds (14,500 kilos) of cargo, mostly contact lenses. According to records, there were no hazardous materials aboard. According to CNN, the flight crew consisted of a Captain with over 7,000 hours, rated for the 767 since May 2026, and a first officer with over 2,500 hours, rated for the 767 since April 2025.

The flight departed San Juan at 15:42 UTC as 21 Air flight 7598 flying for Amazon Prime Air. The routine cargo flight to Miami was just over two hours. The weather at Miami was bad, thunderstorms and gusting winds.

The initial CVR and FDR data released by the NTSB show an unstable approach in the last moments before touchdown.

Six minutes before the end of the CVR recording, one of the pilots called for “flaps 1”, extending the flaps slightly as they began configuring the aircraft for landing. They switched frequencies to Miami Tower, who cleared the aircraft to land on runway 30.

The flight crew extended the flaps further and lowered the landing gear.

One pilot called for flaps to be set to 20°. The other cautioned that they were travelling too fast. This pilot, presumably the Pilot Monitoring, made repeated references to the aircraft’s excessive speed.

The NTSB note only that there was no “consistent verbal response”; I’m not sure if that means there was any acknowledgement of the issue at all. At any rate, the other pilot continued with the landing.

In the cockpit was the sound of the autopilot disconnecting, followed by electronic callouts, a mixture of their height (feet above the runway) and warnings of their excessive descent:

ONE THOUSAND

SINK RATE SINK RATE

FIVE HUNDRED

SINK RATE SINK RATE

ONE HUNDRED

MINIMUMS

TOO LOW TERRAIN TOO LOW TERRAIN TOO LOW TERRAIN TOO LOW TERRAIN

A pilot called for flaps to be set to 30°, the normal landing configuration. It was ten seconds before touchdown.

FORTY

TOO LOW TERRAIN

TEN

The landing was hard and uneven. The FDR shows that the nose gear and right main gear touched down at a groundspeed of 158 knots. Seven seconds later, the brakes were applied. A few seconds after that, the left main gear touched down at a groundspeed of 134 knots.

One of the pilots called for a go-around. Sixteen seconds had elapsed since the initial touchdown.

Still travelling at 120 knots, the flight crew released the brakes and advanced the throttles to go-around power for taking off again.

Four seconds later, the throttles were reduced to idle power and the brakes applied again. The FDR recorded the groundspeed as 117 knots. The brakes remained applied to the end of the FDR recording.

The Boeing 767 ran off the runway and onto an airport service road. It was travelling 96 knots (178 km/hour) when it struck a van on the service road, killing five on board with another two seriously injured. The aircraft continued, crashing through the fence and concrete vehicle barriers before slamming into an SUV, critically injuring one occupant.

When the CVR and FDR stopped recording, the aircraft was still travelling at 65 knots (120 km/h). The aircraft came to rest near an Amazon warehouse parking lot.

The NTSB have released a seven-minute video of their initial on-site response. Please note that this includes multiple views of the crash site.

Flightradar24 have created a clear graphic showing the accident approach compared to four other flights that landed in the hours before, including another Boeing 767.

There’s another aspect of this accident: airport infrastructure. The inescapable fact is that an aircraft landing on a runway is a large and heavy thing with a large amount of momentum that needs to slow down as quickly as is safely possible.

Since the 1980s, the FAA design standard for a runway safety area is 1,000 feet beyond the runway end. Miami International does not have a clear 1,000 feet beyond the physical end of runway 30 nor do they have the land to make one. If an existing airport cannot physically include a thousand-foot runway safety area, they have two choices.

One is to install an Engineered Material Arresting System (EMAS). This is a bed of lightweight, crushable cellular cement blocks installed after the runway’s end. The aircraft’s wheels sink into the material allowing it to decelerate much more quickly. A member of the FAA team described it as “like a runaway truck stop, but for airplanes”. This is the solution at Boston Logan, for example, which is hemmed in by water.

The other option is to truncate the usable landing distances on the runway, that is, to pretend that the runway is shorter than it is in order to allow for part of the runway to act as a runway safety area.

Miami International Airport decided to do the latter. Miami’s runway 30 is 9,360 feet long with a displaced threshold of 945 feet. The published Landing Distance Available (LDA) is 7,913 feet. That allows for 500 feet to be reserved as a safety area buffer which, combined with the terrain beyond the runway, make for a 1,000 foot runway safety area.

On a dry runway, the Boeing 767 requires a runway length of 5,800 to 6,200 feet. Even if the runway was wet from the weather, 7,913 feet should be more than adequate.

But a runway overrun is by definition not normal circumstances and in urban areas, there is an argument that airports need to ensure that aircraft will stop before the end of the safety area. A standard EMAS installation will stop most aircraft travelling at 70 knots off the end of the runway. According to the FAA, as of August this year, 26 overrunning aircraft, carrying 497 crew and passengers were safely stopped through EMAS installations at the end of runways.

Mary Schiavo, former inspector general of the Department of Transportation, argues that EMAS should be required at urban airports.

An airport this surrounded by population, they need an arrestor bed system at the end of these runways to stop this from happening again.

It is very rare that a runway overrun causes deaths; Wikipedia’s attempt at an exhaustive list of notable runway excursions lists 53 that have resulted in “fatalities, aircraft destruction or substantial aviation safety changes” between 1958 and 2026, of which 36 included fatalities, of which only ten included fatalities on the ground. Over the course of almost 70 years, that’s really not aviation’s most pressing problem. After a quick glance over the list, I would say that only six fatal accidents could have been avoided with EMAS. Two are evidence of the success of EMAS (successful stop, no deaths).

The issue is that EMAS only helps in a straight-ahead overrun. Schiavo’s argument implies that those on the ground need more protection than those in the aircraft, whereas I would say that if EMAS is required for specific airports, it should be any airport where buildings, roads or terrain cause a specific risk. That said, I’m not against the idea that airports should be pressed a bit harder to spend the once-per-runway cost to ensure that their aircraft cannot accidentally hit escape velocity and cause a mass casualty event.