The Miami 767 Overrun: Too Fast, Too Long, and a Late Decision
Figures below come from the NTSB's September 9 investigative update and its October 1 preliminary report (DCA26MA352), unless noted. Both are preliminary and subject to change, and the NTSB has not determined a probable cause.
On 6 September 2026, 21 Air Flight 7598, a Boeing 767-33A freighter (N1997A) flying for Amazon from San Juan to Miami, overran the end of Runway 30 at Miami International Airport and struck two vehicles. Five people on the ground were killed, and three were seriously injured. Both pilots had minor injuries. The airplane was substantially damaged, with a post-impact fire on the right wing (National Transportation Safety Board [NTSB], 2026b). Probable cause awaits the final report. But the recorder data already describe a high-energy approach that never became a stable landing, and a go-around attempt that began well into the landing roll.
Flight 7598 was cleared to land on Runway 30 at about 1349 EDT, roughly 12 nautical miles out. Thunderstorms were reported at the field, with visibility of 10 miles or better. The 1353 routine weather report gave wind from 190° at 17 knots gusting to 26, and the peak wind of the hour, 26 knots from 180°, came a minute earlier (NTSB, 2026b). Resolved against the runway heading, that wind is mostly a right crosswind in the mid-teens to mid-20s, with a tailwind component of roughly 4 to 10 knots, depending on which observation and gust you use. That is my arithmetic, not the NTSB's; the report quantifies neither component.
Runway 30 is 9,360 feet long on paper, but its displaced threshold brings the published landing distance available (LDA) down to 7,913 feet (Federal Aviation Administration [FAA], 2026). Runways 26L and 27 publish LDAs of 10,220 and 12,755 feet. The preliminary report does not explain how Runway 30 was assigned or whether a longer runway was considered or requested.
The cockpit voice recorder (CVR) timeline is measured backward from the end of the recording. About four minutes and eighteen seconds out, the crew checked in with Miami tower and was cleared to land. At 1:42, the captain called for flaps 20, and the first officer cautioned that the airplane was too fast. The first officer kept commenting on the excess speed for the rest of the recording, and the NTSB notes the comments did not draw a consistent verbal response. The autopilot appears to have disconnected at 1:23. In the final minute, the recorder captured "sink rate" callouts, a "minimums" callout, and five "too low terrain" alerts; the report does not say which ground-proximity mode produced them. The captain called for flaps 30 at 0:41, within a second of the radio altimeter's "forty" callout and about ten seconds before the sounds of touchdown (NTSB, 2026b). At the wreckage, the flap lever was in the 25° detent, and the slat tracks and flap support rods measured consistent with 25°. The report does not state the airplane's setting at touchdown or why the lever was found short of the commanded setting.
The airplane crossed the Runway 30 threshold about 105 feet above the ground at a groundspeed of 177 knots. The first main-gear tire marks were found 3,750 feet (left) and 3,820 feet (right) beyond the threshold, at about 160 knots, roughly 13 seconds after threshold crossing. The weight-on-wheels signal then flickered between air and ground for about seven seconds, and settled on ground with about 2,730 feet of runway remaining (NTSB, 2026b). The September 9 update, drawn from the first flight data recorder (FDR) readout, described the nose and right main gear touching down at 158 knots, brakes applied at 146 knots, and the left main gear touching down 11 seconds after the right, at 134 knots (NTSB, 2026a). The two NTSB descriptions differ, and the NTSB says an airplane performance study will refine the sequence.
The FDR shows the speed brakes stayed stowed and the thrust reversers never deployed. The crew applied the wheel brakes manually at 146 knots. About eight seconds later, at 120 knots with about 930 feet remaining, the brake pedals were released, and both throttles advanced to go-around thrust. The CVR logs the captain's go-around call about 16 seconds after the touchdown sounds, though the NTSB cautions that the CVR and FDR clocks are not precisely synchronized. About four seconds after the throttle advance, at 117 knots with about 415 feet remaining, the throttles returned to idle, and the brakes were reapplied. The airplane left the pavement at 113 knots and hit the localizer and approach lighting at about 100. It then struck a Ford Econoline van on the airport perimeter road, went through the perimeter fence, crossed NW 67th Avenue, struck a Toyota SUV, and stopped about 1,200 feet past the end of the pavement after hitting a parking-lot fence. Five of the van's seven occupants were killed; the other two, and the SUV's driver, were seriously injured (NTSB, 2026b).
Let’s start with the energy. Crossing the threshold at 105 feet and 177 knots groundspeed is about twice the height of a normal threshold crossing and far outside a stabilized approach. The report gives neither landing weight nor reference speed, so I can’t know the exact overspeed, but the NTSB's performance study will. The airplane flew 3,750 feet down a runway with 7,913 feet of landing distance available before the main gear marked the pavement, and by the time weight-on-wheels settled, 2,730 feet of pavement remained.
I started my airline career in the 1980s, when no real stabilized approach concept existed. You flew a good approach because the captain beside you expected one, and cockpit-to-cockpit handed-down standards defined what counted as good. The formal standard came later. The Flight Safety Foundation formed its Approach-and-Landing Accident Reduction (ALAR) Task Force in 1996 and released the ALAR Tool Kit in 2000 (Flight Safety Foundation [FSF], n.d.). Its briefing note on the stabilized approach recommends being stabilized by 1,000 feet above airport elevation in instrument conditions and 500 feet in visual conditions. Stabilized means on the correct flight path, in the landing configuration, at no less than reference speed and no more than reference plus 20 knots, with a sink rate no greater than 1,000 feet per minute and power appropriate to the configuration. If the approach is not stable at the gate, or becomes unstable below it, the note calls for a go-around (FSF, 2009). That has been the industry standard for a long time now, and for good reason.
Measure this approach against it. The field conditions were reported as VFR, so the gate was 500 feet. The CVR's "five hundred" callout comes 58 seconds before the end of the recording. By then, the last flap setting commanded was 20, which is not a landing setting. The flaps-30 call did not come for about 17 more seconds, within a second of the "forty" callout. The first officer had been calling excessive speed for 44 seconds. The ground-proximity system had already called "sink rate" at 1:04 and called it again at 0:57 (NTSB, 2026b). The report gives no numbers for speed or sink rate at that point, and it does not publish 21 Air's own criteria. But configuration, speed, and sink rate all point to the same conclusion, and I will say plainly what I think: this approach was so grossly unstable that any operator's criteria would have mandated a go-around. The decision that mattered was not the one made with 930 feet of runway left. It was available at 500 feet, and at every point after that down to the flare.
The deceleration devices are the other half of the story. On a heavy jet at this speed, spoilers and reversers carry much of the early deceleration, and neither was used. At the wreckage, the speed brake lever was in the down (stowed) position. The report does not say whether the speed brakes were armed on the approach, which will be a key question. The throttles were at idle, and the reverser levers were in the first detent, which the report does not explain, given the FDR's showing no deployment. Wheel braking did not begin until 146 knots, about seven seconds after first gear contact. A go-around that begins at 120 knots with 930 feet of runway left, after wheel brakes have been on for eight seconds, has very little chance of working.
The captain was the pilot flying and the first officer the pilot monitoring, so the monitoring pilot called out the speed, and the NTSB notes they did not get a consistent response. The experience gradient runs the opposite way from the usual one. The captain, hired in April 2026 and type-rated on the 767 on May 18, reported about 7,267 total hours, including about 2,700 on the 737, but only about 150 in the 767. The first officer, hired in February 2025, reported about 2,800 total hours and about 800 in the 767 (NTSB, 2026b). Those figures come from the pilots' own interviews. This was the second flight the two had flown together; they met that morning on the Miami–San Juan leg, which the first officer flew. The report lays out recent duty history without drawing conclusions. The captain's previous day was a Miami–Bogotá round trip whose return leg departed Bogotá about 0441 and arrived in Miami about 0902, and both pilots reported for duty about 0530 on the day of the accident. Flight 7598 departed San Juan about 1124 and was scheduled to arrive about 1428; the accident occurred about 1353.
Several lines of inquiry do not depend on a probable-cause finding. First, the local controller issued the landing clearance without wind information, although FAA Order JO 7110.65BB requires both direction and velocity when a tailwind component exists. The controller, with 11 years of experience and about 17 minutes on position, had given another flight winds of 190° at 19 knots gusting to 22 about four minutes earlier, when Flight 7598 was not yet on the frequency (NTSB, 2026b). The procedural requirement is clear. Whether the omission affected the crew's decisions is a causation question. A wind call would have told them about gusts and a tailwind component, and whether it would have changed what they did is for the investigation.
Second, the crew pairing. The Part 121 pairing rule requires only that one pilot in the pair have at least 75 hours of line operating time in type (14 C.F.R. § 121.438(b)), and the first officer's reported 800 hours in type would appear to satisfy it on its face. The more productive questions go to training, operating experience, and the operator's own policies on pairing and stabilized approaches, none of which the preliminary report addresses. For completeness of the public record, 21 Air petitioned the FAA in 2025 for relief from §§ 121.438(b) and 121.652(a) (FAA, 2025). I did not determine whether relief was granted, and nothing in the preliminary report suggests it bears on this flight.
Third, the runway end. The NTSB notes that no Engineered Materials Arresting System (EMAS) was installed at the end of Runway 30. But the FAA's design standard for a standard EMAS assumes an aircraft exiting the runway at 70 knots (FAA, 2012), and this airplane left the pavement at 113 knots and struck the localizer at about 100. What an arresting bed would have done at that speed is an open engineering question. A closer question is what lay beyond the runway end: a localizer, approach lighting, a perimeter road, and a fence, all in the airplane's path.
Fourth, the people on the ground. The van was on an airport perimeter road, inside the perimeter fence. Only the SUV was on a public road. The report does not say who the van's occupants were or why they were there, and that will matter to the analysis. A wrongful death suit has already been reported against Amazon, 21 Air, and several leasing entities, and reportedly names the two pilots individually (CBS12, 2026). The 45-day bar on unsolicited attorney contact with victims and families under 49 U.S.C. § 1136(g)(2) ran through October 21 (NTSB, 2026c).
Finally, a note on evidence. The Board's eventual probable-cause report cannot be admitted or used in a civil damages action (49 U.S.C. § 1154(b); 49 C.F.R. § 835.2), but the regulation states that there is no statutory bar to factual accident reports, including group chairman factual reports. The litigation will turn on the recorder data and group reports.
The preliminary report leaves much unanswered. It does not say why the speed brakes and reversers were not used, or whether the speed brakes were armed. It does not say what flap setting the airplane had at touchdown. It does not give the landing weight and reference speed, which the performance study will. It does not reconcile the tire-mark data with the gear-by-gear touchdown description. It does not describe 21 Air's stabilized-approach and go-around criteria, or explain how Runway 30 was assigned. It also draws no conclusions about crew duty history or fatigue.
What the record shows so far is an airplane that arrived over the threshold high and fast, touched down nearly halfway down the available landing distance, rolled most of the remaining pavement without spoilers or reversers, and began a go-around only when the runway was almost gone, while required wind information had not been given. Those facts are enough for operational lessons now. As for the legal analysis, that has already begun.
---
References
CBS12. (2026). Wrongful death lawsuit filed against Amazon over deadly Miami cargo jet crash [News report]. https://wjla.com/news/nation-world/wrongful-death-lawsuit-filed-against-amazon-over-deadly-miami-cargo-jet-crash-amazon-prime-air-flight-7598-miami-international-airport-miami-cargo-plane-crash-miami-plane-crash-wrongful-death-lawsuit-amazon-lawsuit-21-air-atlas-air-worldwide
Federal Aviation Administration. (2012, September). Engineered materials arresting systems (EMAS) for aircraft overruns (Advisory Circular 150/5220-22B). https://www.faa.gov/documentLibrary/media/Advisory_Circular/150_5220_22b.pdf
Federal Aviation Administration. (2025). Petition for exemption; summary of petition received; 21 Air Inc. (Summary Notice No. 2025-38; Docket No. FAA-2025-0464). https://public-inspection.federalregister.gov/2025-10523.pdf
Federal Aviation Administration. (2026). Chart supplement: Southeast U.S. (p. 142, Miami International Airport). https://airportdocs.iflightplanner.com/AFD/260709/se_142.pdf
Flight Safety Foundation. (n.d.). Partners and programs [80 years of aviation safety series]. https://flightsafety.org/asw-article/80-years-of-aviation-safety-6-2/
Flight Safety Foundation. (2009). ALAR briefing note 7.1: Stabilized approach. https://skybrary.aero/sites/default/files/bookshelf/864.pdf
National Transportation Safety Board. (2026a, September 9). NTSB issues investigative update on B-767 runway excursion accident in Miami [News release]. https://www.ntsb.gov/news/press-releases/Pages/NR20260909.aspx
National Transportation Safety Board. (2026b, October 1). Aviation investigation preliminary report: Runway overrun of 21 Air flight 7598, Boeing 767-33A, N1997A, Miami, Florida (DCA26MA352). https://www.ntsb.gov/investigations/Documents/DCA26MA352%20Prelim.pdf
National Transportation Safety Board. (2026c, October 1). Runway overrun of 21 Air flight 7598 [Investigation page]. https://www.ntsb.gov/investigations/Pages/DCA26MA352.aspx
14 C.F.R. § 121.438 (2026). https://www.law.cornell.edu/cfr/text/14/121.438
49 C.F.R. § 835.2 (2026). https://www.ecfr.gov/current/title-49/subtitle-B/chapter-VIII/part-835/section-835.2
49 U.S.C. § 1136(g)(2).
49 U.S.C. § 1154(b).
