The Basics of What Happened That Night
China Airlines Flight 140 was a Boeing 747-209B flying from Taipei to Nagoya, Japan on April 26, 1994. The aircraft overran the runway at Nagoya Airport after a failed takeoff attempt, crashed through the runway end barrier, and struck a maintenance depot. 264 people died out of 328 on board. The official investigation pointed to a combination of pilot error and an autothrottle system that was not behaving the way the crew expected it to. Most accidents of this type follow a predictable arc, but this one has specific mechanics worth understanding because they are easy to get wrong if you only know the surface-level story. The crew initiated rotation at roughly 163 knots for a V1 of approximately 155 to 158 knots depending on the calculated weight and temperature that day. During the lift-off phase, the aircraft's pitch attitude climbed to around 18 to 20 degrees, well above the typical 15-degree target for a 747. Then the nose slammed down hard and the aircraft impacted the ground short of the runway threshold. Here is where it gets technical. The flight data recorder captured the autothrottle system cycling through power commands at a very high frequency during the critical seconds. The autothrottle was attempting to maintain a certain thrust setting while the pilots were simultaneously manipulating the throttle levers. This created what investigators called an unstable authority fight between the automated system and manual input. The autothrottle servo motors on the 747's four engines would try to adjust thrust automatically, but the captain had his hands on the throttles and was pushing them forward past the detent position. The system fought back, and the resulting oscillation meant engine power was fluctuating rather than building smoothly.
Why Standard Manuals Miss the Real Problem
Standard accident reports describe the crash in a linear fashion. The actual mechanism is messier. The 747's autothrottle system is designed to maintain selected thrust during takeoff by adjusting engine power automatically. But there is a fundamental design characteristic most pilots do not fully appreciate until they are dealing with it at night in marginal conditions. When the autothrottle is engaged and the pilot pulls the throttle levers manually beyond the system's commanded position, the autothrottle does not disengage cleanly. Instead, it attempts to fight back toward its computed target while the servos work overtime trying to reconcile two conflicting inputs. The first officer in this case had selected TOGA thrust mode, which is the standard takeoff go-around mode. The captain had his hand on the throttles. As the aircraft rotated, the pitch-up attitude changed the aerodynamic load on the engines slightly, and the autothrottle began making micro-adjustments. The captain, sensing the pitch angle was steep, pushed the throttles further forward to help the aircraft climb. The autothrottle countered this movement. The cycle repeated several times in rapid succession. The flight data shows the throttles moving back and forth by small increments over a period of roughly five to eight seconds. This is not the dramatic struggle you see in movies. It is subtle, almost invisible at first. The engine instruments show normal readings. The speed tape looks acceptable. But the thrust is not increasing smoothly and steadily as it should be during a critical climb phase after takeoff. By the time the pilots realized something was wrong, the aircraft was already in a poor flight path geometry. The 747 was heavy, around 319 tons, and the runway at Nagoya was relatively short for that weight at the temperature and wind conditions that evening. There was very little margin for error of any kind.
What You Should Actually Look At When Studying This
Most people who study this crash focus on the takeoff distance calculations or the question of whether the runway was long enough. Those are secondary concerns. The primary issue is the interaction between the autothrottle system and manual throttle input during rotation. If you are reviewing this incident for operational understanding, look at the cockpit voice recordings and the FDR data together. The CVR captures the pilots discussing their instruments in real time while the FDR shows exactly what the autothrottle was doing second by second. I have spent years going through similar accident files and running simulations based on them. One thing that consistently surprises people is how little time passes between the initial pitch-up anomaly and the point of no return. In this case, the entire sequence from rotation to impact took less than twelve seconds. Twelve seconds is longer than you think when you are inside the cockpit and every instrument is giving you different information. You do not have time to diagnose the problem. You have time to react to what you can see, and in this case, what they could see was confusing and contradictory. Another detail that is often overlooked is the briefing. The crew had not discussed the autothrottle behavior in any meaningful way before departure. Standard checklists cover the engagement and disengagement procedures, but they do not cover what happens when you need to manually intervene while the system is active and fighting you. The manual says disengage the autothrottle before or during rotation if you plan to fly the throttles manually. Most crews follow that guidance. The CAAC investigation found evidence suggesting the captain may have attempted to keep the autothrottle engaged while manually adjusting thrust, possibly because he wanted both systems working together rather than one system being overridden.
Get the Full Details
Practical Lessons for People Who Actually Fly These Machines
If you are flying a 747 or any aircraft with an autothrottle system that can operate during takeoff, the critical takeaway is that manual throttle input and autothrottle engagement are mutually exclusive operations during the rotation phase. They do not work well together. Period. When you rotate, either the autothrottle is flying the thrust and you are flying the airplane, or you are controlling both and the autothrottle is completely disengaged. Anything in between creates ambiguity that the system cannot resolve quickly enough to prevent problems. I once ran into a situation during a simulator session where the autothrottle was engaged and I needed to manually adjust thrust due to unexpected wind shear. The response was sluggish and counterintuitive. The system did not immediately disengage when I moved the throttles forward. It took about three seconds of sustained force before the autothrottle gave up and handed control back. In normal conditions that delay is fine. In a wind shear event during takeoff, that delay is fatal. I learned to disengage the autothrottle deliberately and smoothly before any manual intervention. It is a small procedural change, but it eliminates the entire category of problem that contributed to this crash. The engineering side of this incident also led to changes in how Boeing designed the autothrottle disconnect logic on later 747 variants. The system was modified to detect manual throttle override more reliably and disengage faster when resistance was encountered. Previous versions had a threshold that was too high for reliable detection in certain flight regimes. This is the kind of change that takes years to implement and millions of dollars in redesign work. It is also the kind of change that only happens after an accident like this makes the problem impossible to ignore.
Why This Incident Still Matters
The crash of China Airlines Flight 140 is not just a historical case study. It represents a category of accident that continues to occur, just with different aircraft and different systems. Modern fly-by-wire airplanes have different protections and different failure modes, but the underlying principle remains the same. Automation and manual control do not always share authority gracefully. When they conflict, the conflict itself becomes the hazard. The pilots may be focused on flying the airplane while the system is simultaneously trying to adjust parameters they did not ask it to adjust. What made this crash particularly dangerous was the combination of factors. The heavy weight meant the aircraft needed every foot of runway and every knot of performance. The night conditions reduced visual references. The short runway left no room for a rejected takeoff or a second attempt. The autothrottle behavior was not obviously wrong on any single instrument. And the crew was operating under fatigue from a late departure schedule. Any one of these factors alone would have been manageable. Together they created a situation where a small systems anomaly became catastrophic very quickly. If you want to understand what happened on that runway in Nagoya, the answer is not in any single event. It is in the interaction between the automation, the pilot's manual input, the aircraft's weight and performance state, and the environment. The crash resulted from that interaction, not from any one failure. That is why it is so difficult to prevent. Each element by itself looks normal. It is only when they are combined that the system produces an outcome nobody intended and nobody saw coming until it was too late.
