Space is big, but orbits are narrow. Screening systems flag thousands of conjunctions every day - moments when two objects are predicted to pass uncomfortably close - and nearly all of them end in misses measured in kilometers. This list is about the other kind. The passes where the error bars overlapped the hardware, where operators burned propellant with hours to spare, or where nobody could do anything but watch.

A word about the numbers before we rank anything. A conjunction warning is a prediction, not a measurement. It comes from radar and optical tracks with uncertainties that can span tens to hundreds of meters, so “predicted miss distance of 12 meters” really describes the center of a probability cloud. That’s why operators talk about collision probability instead of distance, and why a scary headline number can evaporate with the next tracking pass. Our explainer on how conjunction screening works covers the machinery; below, we label what was predicted and what was measured.

Some ground rules for this list. We rank by how close the pass really was and how bad a hit would have been, using the best post-event numbers available. Predictions count, but measurements count more. And the list is bookended by the two times cataloged objects actually connected, because that’s the outcome space situational awareness exists to prevent.

Most of the objects involved are still up there, and you can pull any of them up in KeepTrack.

1. Iridium 33 vs Cosmos 2251

Date: February 10, 2009 | Objects: Iridium 33 (active) vs Cosmos 2251 (defunct) | Predicted miss: 584 m | Outcome: hit

The list starts with the day the odds ran out. An operational Iridium communications satellite and a dead Russian Strela-2M relay crossed paths 789 kilometers over Siberia at 11.7 kilometers per second, and this time there was no miss. Both spacecraft were destroyed instantly, and the wreckage, nearly 2,300 cataloged fragments, remains a dominant debris hazard in that altitude band today.

The bitter detail is that this conjunction had been screened. CelesTrak’s SOCRATES service predicted a 584-meter miss, unremarkable for that week. Iridium received warnings like it constantly and couldn’t burn fuel against all of them. The full story of the collision became the founding case study for every screening system built since. Each entry below is a rerun of this setup that happened to end differently.

2. TIMED vs Cosmos 2221

Date: February 28, 2024 | Objects: TIMED (active, no propulsion) vs Cosmos 2221 (defunct) | Measured miss: under 10 m | Outcome: no maneuver possible

NASA’s TIMED spacecraft has studied the upper atmosphere since 2001, and it carries no propulsion. So when tracking showed it converging with Cosmos 2221, a defunct Russian electronic-intelligence satellite, at about 608 kilometers altitude, all anyone could do was refine the estimate. They crossed at 1:34 a.m. Eastern at a closing speed around 28,000 kilometers per hour.

The initial word was a pass of roughly 20 meters. Weeks later, NASA Deputy Administrator Pam Melroy revealed that reconstructed tracking put the miss at under 10 meters - about the length of the spacecraft themselves. A hit would have shredded both vehicles at an altitude shared by hundreds of active satellites. It’s the closest documented miss between two intact satellites, period.

Track TIMED live in KeepTrack, and Cosmos 2221 as well - the pair will keep crossing paths until their orbits drift apart.

3. IRAS vs GGSE 4

Date: January 29, 2020 | Objects: two defunct satellites | Predicted miss: 12-47 m | Outcome: no maneuver possible

IRAS mapped the infrared sky for ten months in 1983 and has been a one-tonne derelict ever since. GGSE 4 is a Naval Research Laboratory experiment from 1967 that flew with the classified POPPY signals-intelligence satellites. The two converged 900 kilometers above Pittsburgh at 14.7 kilometers per second while LeoLabs radars watched the predicted miss bounce between 12 and 47 meters over the final days.

Both satellites passed clean, confirmed by radar minutes later. Dead-versus-dead conjunctions are the category trackers lose sleep over, since nobody can move either object. The only options are watching and math.

See IRAS in KeepTrack, or find GGSE 4, boom and all.

4. Cosmos 2004 vs a CZ-4C rocket stage

Date: October 16, 2020 | Objects: defunct Soviet satellite vs spent Chinese upper stage | Predicted miss: 25 m ± 18 m | Outcome: no maneuver possible

Nine months after the Pittsburgh scare, the same radar network flagged a heavier pairing. A dead Soviet navigation satellite and the spent upper stage of a Chinese CZ-4C rocket, combined mass around 2,800 kilograms, were converging at 991 kilometers over the Weddell Sea. The final estimate gave a 25-meter miss with 18-meter error bars and collision odds above 10 percent, among the highest figures ever published for two large intact objects.

At 991 kilometers there is no atmospheric cleanup on any human timescale. A hit would have salted one of the most crowded altitude bands in orbit with debris for centuries. Ten minutes after the predicted impact time, the rocket body swept over a radar in New Zealand as a single intact object.

See Cosmos 2004 in KeepTrack - the rocket stage is in the catalog too.

5. The Canadarm2 debris strike

Date: discovered May 12, 2021 | Objects: ISS robotic arm vs untracked debris | Predicted/measured miss: 0 m (impact) | Outcome: hit, minor damage

Nobody saw this one coming, because nobody could. During a routine inspection, cameras found a 5-millimeter hole punched clean through the thermal blanket and into the boom of Canadarm2, the station’s 17.6-meter robotic arm. The Canadian Space Agency called it a “lucky strike,” and given the arm is only 35 centimeters across, the phrase fits. The impactor was never tracked before the hit and was never found after it.

Conjunction screening defends against the cataloged population, objects roughly softball-sized and up. Below that size circle millions of fragments capable of doing this or worse. The arm still works fine. The hole stays.

6. The ISS and the Cosmos 1408 cloud

Date: November 15, 2021 | Objects: ISS vs fresh anti-satellite-test debris | Predicted/measured miss: repeated debris-field crossings | Outcome: crew sheltered

When Russia destroyed its own Cosmos 1408 satellite with a ground-launched interceptor, the roughly 1,500 trackable fragments appeared almost on top of the International Space Station’s orbit. Seven crew members were woken with instructions to seal hatches and shelter in their Dragon and Soyuz lifeboats as the station crossed near the fresh cloud roughly every 90 minutes.

The station has performed more than 40 debris avoidance maneuvers since 1999, and in the years since the test, several have been against pieces of Cosmos 1408. Most conjunctions on this list involve two objects. This one involved one station and a shotgun blast.

Track the ISS in KeepTrack.

7. Aeolus vs Starlink 44

Date: September 2, 2019 | Objects: ESA’s Aeolus vs Starlink 44 | Predicted odds: ~1 in 1,000 | Outcome: dodged

This entry earns its place through process failure, not distance. ESA’s wind-mapping satellite Aeolus and one of the first sixty Starlink satellites were converging with collision odds around 1 in 1,000, ten times ESA’s maneuver threshold. ESA emailed SpaceX. SpaceX replied early on that it had no plans to move, and then a bug in its on-call paging system meant nobody there saw the risk climb afterward.

So Aeolus moved, firing thrusters three times to raise its orbit about 350 meters, the first collision-avoidance maneuver ever flown against a mega-constellation satellite. The lasting fix wasn’t better email. Both organizations pushed toward automated coordination, because hand-negotiating one conjunction at a time doesn’t scale to tens of thousands of satellites. Aeolus was deliberately deorbited in 2023.

8. Tiangong vs Starlink, twice

Date: July 1 and October 21, 2021 | Objects: Tiangong station (crewed) vs Starlink-1095 and Starlink-2305 | Predicted/measured miss: not disclosed | Outcome: dodged twice

In a note verbale to the UN Secretary-General that December, China reported that its crewed Tiangong station had maneuvered twice in 2021 to avoid Starlink satellites. The October case involved a Starlink that was actively raising its orbit, which China said made the trajectory hard to predict. The US replied that neither encounter met emergency thresholds and that nobody had contacted US Space Command or SpaceX about them beforehand.

Both claims can be true at once. The episode exposed the lack of any working coordination channel between the two governments whose hardware dominates low Earth orbit. Diplomatic notes are not a conjunction-screening protocol.

See Tianhe, Tiangong’s core module, in KeepTrack.

9. OneWeb-0178 vs Starlink-1546

Date: April 3, 2021 | Objects: OneWeb-0178 vs Starlink-1546 | Predicted miss: 58 m claimed; ~1,100 m measured | Outcome: dodged, then disputed

This is the list’s cautionary tale about headlines. Early Space Force screening data days before the pass suggested the two broadband satellites could come within about 58 meters, with collision odds of 1.3 percent, and OneWeb executives took that story to the press. SpaceX called the account misleading, and the post-event tracking backed it up. Space Force data put the actual closest approach at 1,120 meters; LeoLabs independently measured 1,072.

An estimate made days out carries enormous uncertainty that shrinks as fresh observations arrive, which is why nobody serious judges a conjunction by a single early miss-distance number. OneWeb did maneuver, coordination between the companies did improve afterward, and “58 meters” lives on in aggregator posts to this day.

Find OneWeb-0178 in KeepTrack.

10. Cerise vs an Ariane fragment

Date: July 24, 1996 | Objects: Cerise vs Ariane 1 stage fragment | Predicted/measured miss: 0 m (impact) | Outcome: hit, satellite survived

The list closes where the whole subject began. Cerise, a year-old French military signals-intelligence microsatellite, was hit at around 14 kilometers per second by a cataloged fragment of an Ariane 1 upper stage that had exploded back in 1986. The impact sliced off 4.2 meters of the satellite’s gravity-gradient stabilization boom and sent it tumbling. It was the first confirmed accidental collision between two cataloged objects in orbit.

French engineers pulled off a save, rewriting the attitude-control software to fly the wounded satellite without its boom and resuming the mission. Cerise proved the thing Donald Kessler had warned about two decades earlier, that debris begets debris and yesterday’s explosion becomes tomorrow’s bullet. Thirteen years later, entry number one proved it at full scale.

See Cerise in KeepTrack - still in orbit, scar and all.

The ones that just missed

Two honorable mentions from the actually-hit category. In 2016, ESA’s Sentinel-1A took a millimeter-scale debris strike that gouged a 40-centimeter damage zone in a solar array, photographed by the satellite’s own onboard cameras. And in 2013, BLITS, a small Russian laser-ranging sphere, abruptly changed spin and orbit after a suspected hit from a fragment of China’s 2007 anti-satellite test. Small impacts happen all the time. These two just happened where instruments could confirm them.

The scoreboard runs continuously

Close calls stopped being rare events and became a background hum. SpaceX reported 207,152 Starlink collision-avoidance maneuvers in the six months ending May 2026, roughly 355,000 over the full year, and screening duties for US civil operators are moving to the Office of Space Commerce’s TraCSS system. None of those maneuvers made headlines, which is the point.

You can watch the raw material for yourself. Open KeepTrack and look at the 900-1,000 kilometer shell where entries 3 and 4 nearly happened - the density on screen is the conjunction problem, live. And since most of this list’s cast are dead satellites that can’t dodge anything, our tour of the 10 zombie satellites makes a natural next stop.