How often orbital launches fail — the number of failures over the last 365 days as a share of every launch attempt worldwide in that window.
Series begins Dec 2015 — the first month with a complete 12-month window in our backfilled history.
Highest failure rates among launch providers with ≥5 launches in the last 24 months.
19 launch providers excluded — fewer than 5 launches in the last 24 months. A single failure on a small sample would dominate the chart.
Highest failure rates among vehicles with ≥5 launches in the last 24 months.
62 vehicles excluded — fewer than 5 launches in the last 24 months. A single failure on a small sample would dominate the chart.
What failed, not just that something failed. Editorially curated; counts here come from the failure-event log below.
One of the upper-stage engines did not produce sufficient thrust to reach the planned 460 km circular orbit. AST SpaceMobile's BlueBird 7 satellite was left in a 154 x 494 km elliptical orbit too low for its electric propulsion to recover and will be deorbited.
First flight of a reused New Glenn first stage. FAA temporarily grounded the vehicle pending investigation. Cross-checked against Spaceflight Now and Via Satellite — both corroborate upper-stage engine underperformance.
Approximately 33 seconds into the maiden flight, a small explosion tore through the first-stage engine bay; one of the nine engines over-expanded and ruptured its nozzle. Engine-out capability did not save the vehicle.
No operational payload — maiden flight of reusable booster positioned as a Falcon 9 analog. Specific root cause remains under investigation by Chinese authorities and full technical disclosure is not expected given Chinese commercial-launch transparency norms. Cross-checked against Wikipedia and Orbital Today — both corroborate the engine-bay event.
About 70 seconds after liftoff the vehicle began tumbling during first-stage burn and the autonomous flight safety system was activated, destroying the rocket. Five small satellites were lost.
Underlying cause: Space One did not disclose a technical reason beyond stating mission success was 'difficult'. Third consecutive KAIROS failure. Cross-checked against The Japan Times and Payload — both corroborate first-stage tumble and AFSS activation.
Shortly after liftoff the launch vehicle experienced an anomaly during early first-stage flight, resulting in early termination. Six payloads were lost including the Lilac-3 ultra-flat disk satellite.
Maiden flight of Ceres-2. Galactic Energy confirmed loss but the specific cause has not been publicly disclosed. Cross-checked against Global Times (state media) and China-in-Space — both confirm first-stage anomaly; no technical detail released.
A third-stage anomaly during ascent caused loss of the Shijian-32 satellite. CASC confirmed the failure approximately 12 hours after liftoff.
First outright Long March 3B failure since April 2020 and first full failure across the Long March family in approximately 300 launches. Third-stage cause under continuing investigation. Cross-checked against Xinhua (state media) and China-in-Space — both confirm third-stage anomaly; no technical detail disclosed.
Near the end of the third-stage (PS3) burn, a roll-rate disturbance caused the rocket to spin uncontrollably just prior to stage separation, ending the flight. EOS-N1 and 15 secondary satellites were lost.
Underlying cause: ISRO investigators have pointed to asymmetric side-venting producing torque that overwhelmed the attitude-control system. Second consecutive PSLV third-stage failure following PSLV-C61 (May 2025). Cross-checked against The Print and India TV — both corroborate the third-stage roll-rate anomaly.
About 30 seconds after liftoff the first-stage hybrid motor suffered an anomaly and the vehicle broke into three or four parts, falling within the ground safety zone. Five satellite payloads were lost.
Underlying cause: gas leak in the forward section of the first-stage hybrid combustion chamber triggered a rupture and loss of thrust. Inaugural orbital flight from Brazil's Alcantara Space Center. Cross-checked against SpaceNews and Korea Times — both corroborate the ~30-second anomaly and post-flight investigation findings.
During fairing separation, physical contact or debris damaged the satellite mounting and the second-stage engine's fuel tubing. The second-stage engine's second ignition failed to start normally and shut down prematurely; QZS-5 was placed in the wrong orbit and lost.
JAXA's January 20, 2026 progress report identified payload fairing separation as the root cause. Second H3 failure overall after the maiden flight in March 2023. Cross-checked against JAXA's official press release and SatNews — both confirm fairing-separation root cause.
After successful flight of the first three stages, the fourth and final stage shut down prematurely about 510 seconds into flight. Three payloads were lost: two Jilin-1 Earth-observation satellites and a craft developed by Zhongbei University.
Second Ceres-1 failure in 22 missions (the first was in September 2023). Galactic Energy publicly apologized but did not disclose a root cause. Cross-checked against Space.com and SCMP — both corroborate fourth-stage premature shutdown.
At T+4:18, an electrical arc from the second stage's 450 V DC main power cable shorted out control units, causing loss of engine gimbal control. The flight termination system then detonated the second stage along with its payloads.
Underlying cause: arc discharge from the high-power 450 V DC bus under low atmospheric pressure conditions during second-stage flight. Payloads believed to be Guo Wang internet constellation test satellites though customer not officially disclosed. Cross-checked against China-in-Space and SCMP — both corroborate the electrical-arc cause and FTS activation.
A launch counts as a failure when the rocket fails to deliver its payload to the intended orbit — either a total loss (the vehicle is destroyed or never reaches orbit) or a partial failure (it flies, but the mission is compromised, e.g. the payload is left in the wrong orbit).
The rate is those failures over the trailing 365 days, divided by every orbital launch attempt in the same window. It's based on Launch Library 2's status codes (4 — Launch Failure and 7 — Partial Failure). Pad explosions before liftoff, suborbital tests, scrubbed launches, and cases where the rocket worked but the satellite later malfunctioned on its own are not counted.
Failures sits alongside Launches (the denominator) and Mass to orbit (which counts successful launches only). The trajectory chart on this page reads the same Launch Library 2 records as /launches; the difference is the filter (status 4 + 7) and the rolling 12-month framing.
See When two metrics disagree on the methodology page.
Orbital launch failures over the trailing 365 days, divided by total orbital launch attempts in the same window.
Launch Library 2, /launch/previous/, backfilled into a Supabase historical_launches table for the trajectory and breakdowns
Status code 4 (Launch Failure) and status code 7 (Partial Failure — vehicle launched but mission compromised)
Pad explosions before liftoff, suborbital tests, scrubbed launches, status 9 (Payload Failure) where the vehicle reached orbit but the payload misbehaved separately
Rolling 12-month rate computed monthly from January 2015 to the current month. Each point is failures ÷ launches over the trailing 12 calendar months — chosen as a long enough window to smooth single-event noise while still responsive to recent cadence.
Rate (not count) by launch provider and by vehicle, last 24 months. Sample-size threshold: only entries with ≥5 launches in the window appear. A single failure on a small sample would otherwise dominate the chart. Entries excluded under the threshold are footnoted on each card.
Raw failure counts would put the biggest launchers at the top regardless of reliability — SpaceX would lead the table because it launches the most, not because it’s unreliable. Rates with a sample-size threshold answer the actual question (“who fails most per attempt, given enough attempts to mean something”).
Counts come from a separately-curated failure_events table; each event is tagged with one of a small controlled vocabulary (first-stage anomaly, upper-stage anomaly, stage separation, fairing separation, guidance/control, engine ignition, pad anomaly, range safety, payload deployment, unknown). LL2 doesn’t carry this detail — it tells me a launch failed, not what failed. The vocabulary is fixed in code so cross-event totals are comparable.
Editorial table. Each entry has a primary source URL, source name, source date, and a confidence indicator (confirmed / reported / disputed). I do not republish source articles; I summarise and link. When sources disagree on whether something was a failure, partial failure, or success, the entry is flagged ‘disputed’ with a notes line.
Counts on this page use rolling windows — 12 months for the trajectory, 24 months for the breakdowns — not program lifetime. They will diverge from program-lifetime counts published elsewhere whenever a vehicle’s early flights pre-date the window. SpaceX Starship is the most prominent example: its program record includes Integrated Flight Test 1 (April 2023) and Integrated Flight Test 2 (November 2023) as failures, but both fall outside the trailing 24 months and are not counted here. Launch Library 2 classifies them as failures; the gap is purely windowing, not a classification disagreement.
“Partial failure” is a subjective designation, curated by the LL2 community. Reasonable people would code some borderline cases differently. The trajectory’s history is only as deep as the LL2 backfill; if backfill is incomplete, the chart starts at the first month with a complete 12-month window.
Attribution required