ORIGINAL RESEARCH · 70+ MISSIONS ANALYZED

Why Rockets Fail

An opinionated analysis of orbital launch failures from 1957 to 2025. We catalogued every public failure, tagged the root cause, and found the same seven patterns keep killing missions — decade after decade.

72
Failures Catalogued
7
Recurring Patterns
~30%
Caused by Engine Thrust Loss
68
Years of Data
EDITORIAL · OUR FRAMEWORK

The 7 Failure Patterns

We tagged every failure by root cause. These seven patterns account for nearly every orbital launch failure in history.

PATTERN 01

Engine Thrust Loss

The engine lights, runs for a while, then underperforms or dies. Turbopump seizure, injector erosion, combustion instability, or simple propellant exhaustion.

~30% of all failures
Examples: Falcon 9 CRS-7 (2015), Proton-M/Mexsat (2015), Delta IV GPS anomaly (2012)
PATTERN 02

Guidance & Avionics

The rocket flies perfectly — in the wrong direction. Software bugs, sensor failures, IMU drift, or flight computer crashes send the vehicle off-course.

~18% of all failures
Examples: Ariane 5 Flight 501 (1996), Proton-M/Glonass (2013), Zenit-3SL/Intelsat (2013)
PATTERN 03

Stage Separation

The handoff between stages fails. Explosive bolts misfire, fairings don't jettison, interstage adapters collide with the upper stage, or ullage motors don't fire.

~14% of all failures
Examples: Falcon 1 Flights 1–3 (2006–08), PSLV-D1 (1993), Long March 3B/Intelsat (1996)
PATTERN 04

Structural / Aerodynamic

The vehicle breaks apart under aerodynamic loads, thermal stress, or resonance vibration. Max-Q is the killing field.

~12% of all failures
Examples: Challenger STS-51-L (1986), Antares Orb-3 (2014), N1 launches (1969–72)
PATTERN 05

Propellant / Pressurization

Fuel leaks, helium bottle failures, cryo-loading errors, or pressurization system collapses. The plumbing kills you before the engine ever fires.

~11% of all failures
Examples: SpaceX AMOS-6 (2016), Proton-M/Ekpress (2014), Titan 34D (1986)
PATTERN 06

Human / Process Error

Inverted sensor installation, wrong unit conversion, skipped checklist step, or test data left in flight software. The rocket works as designed — the design was wrong.

~9% of all failures
Examples: Mars Climate Orbiter (1999), Proton-M/Glonass yaw sensor (2013), Ariane 5 reuse of Ariane 4 code (1996)
PATTERN 07

Software-Only Failures

No hardware breaks. A timing bug, integer overflow, race condition, or bad state machine transition kills the mission from the flight computer.

~6% of all failures
Examples: Ariane 5 (64-bit → 16-bit overflow, 1996), Phobos-Grunt (2011), Schiaparelli lander (2016)

Our Take

The uncomfortable truth: the same seven patterns have killed rockets since 1957. New entrants don't fail in novel ways — they rediscover the same failure modes their predecessors hit decades ago. The difference between a company that survives its failures and one that doesn't isn't engineering talent. It's whether they built a culture that treats near-misses as data, not luck.

The Failure Database

Every catalogued orbital launch failure. Search, filter, sort. Click any column header to sort.

Year Rocket Payload Root Cause What Happened
EDITORIAL · DEEP DIVE

Anatomy of a Failure Chain

Rockets don't fail from one thing. They fail from a chain of small things that individually seem harmless.

Case Study: Falcon 9 CRS-7 (June 2015)
A $110M ISS resupply mission lost because of a $2 steel strut.
Strut rated for 10,000 lbf
Actual load exceeded rating
Strut snapped at T+139s
Helium bottle released into LOX tank
Second stage overpressure
Vehicle breakup at 45 km
The lesson that keeps getting relearned: The strut was a commercial off-the-shelf (COTS) part. SpaceX trusted the vendor's rated strength. After the failure, they found some struts failed at just 2,000 lbf — 80% below spec. The fix wasn't a better strut design. It was testing every single strut before flight. The pattern: trust your supply chain, but verify the actual parts, not the paperwork.
Case Study: Ariane 5 Flight 501 (June 1996)
$370M lost in 37 seconds because of a software reuse decision.
Ariane 4 inertial code reused
64-bit float → 16-bit int conversion
Integer overflow at T+37s
Backup computer had same bug
Both guidance computers crashed
Self-destruct triggered
The lesson that keeps getting relearned: The code worked flawlessly on Ariane 4 for years. The variable that overflowed was a horizontal velocity value that Ariane 5's faster trajectory exceeded. Nobody re-validated the assumptions because the code was "flight-proven." The pattern: flight heritage is not the same as flight qualification. Context changed; the code didn't.
EDITORIAL · STRONG POV

Lessons That Keep Getting Relearned

These mistakes have been made, documented, fixed, forgotten, and then made again across decades and companies.

11×
repeated

"We tested the design, not the actual part"

The spec says the strut holds 10,000 lbs. But the actual strut on the vehicle was never tested. Acceptance testing of individual flight hardware catches the supplier defect, the bad weld, the contaminated batch.

CRS-7 2015Antares Orb-3 2014Proton-M 2015Titan 34D 1986
repeated

"The software worked before, so we didn't re-validate"

Code that's "flight-proven" on one vehicle gets reused on a new vehicle with different flight dynamics, different sensors, or different timing requirements. The code is correct — for the old context.

Ariane 5 1996Phobos-Grunt 2011Mars Climate Orbiter 1999Schiaparelli 2016
repeated

"We had the telemetry. Nobody was watching."

Post-failure investigations routinely find that warning signs existed in the data from previous flights. Anomalies were noted, flagged, discussed — then accepted as nominal because the mission succeeded anyway.

Challenger 1986Columbia 2003Proton-M 2013SpaceX B1086 2025
repeated

"The sensor was installed upside down"

It sounds too dumb to be real, but inverted sensors, swapped connectors, and metric/imperial unit mismatches have destroyed rockets worth hundreds of millions. Process and QA failures at the integration level.

Proton-M 2013Mars Climate Orbiter 1999NOAA-19 drop 2003
repeated

"The backup system had the same bug"

Redundancy only works if the backup follows a different failure path. Running the same software on both computers, using the same supplier for both sensors, or routing both cables through the same conduit creates the illusion of redundancy.

Ariane 5 1996Challenger O-rings 1986Phobos-Grunt 2011

Failures by Decade

The rate dropped as the industry matured — then new entrants reset the clock.

1950s
lots
14+
1960s
peak era
20+
1970s
8
1980s
7
1990s
8
2000s
6
2010s
new entrants
9
2020s
5

Why failures went back up in the 2010s

The 2010s saw more new launch vehicle companies than any decade since the 1960s. SpaceX, Rocket Lab, Virgin Orbit, Astra, Firefly — each had to learn the hard way that rockets don't care about your funding round. The 2020s trend is improving, partly because the survivors internalized the lesson: iterate fast, but take anomalies seriously.

Methodology & Sources

This database was compiled from public sources including Encyclopedia Astronautica, Jonathan McDowell's Launch Log, NASA mishap reports, ESA inquiry board findings, and SpaceX public updates. Root-cause categories are our editorial classification based on published investigation findings. Where investigations are still pending or where causes overlap, we used our best judgment and noted it. This is a living page — we update it as new failures occur and as investigation results are published.

Scope: Orbital launch attempts only. Suborbital flights, missile tests, and upper-stage/spacecraft failures after successful orbit insertion are excluded unless the failure directly relates to the launch vehicle. Entries from before 1990 are sampled (not exhaustive) to show pattern continuity.

Last updated: June 2025  ·  Maintained by: Lasting Apps

← Back to Labs