In 1901, sponge divers off the Greek island of Antikythera pulled a corroded lump of bronze out of a Roman shipwreck, and it sat in an Athens museum for half a century before anyone realised they had found a 2,000-year-old computer that could predict eclipses 19 years in advance.

Detailed view of hands crafting a metal component on a lathe. Precision and skill in focus.

In the spring of 1900, a crew of Symi sponge divers en route to fishing grounds off North Africa were forced to shelter from a storm off the island of Antikythera. A diver surfaced babbling about horses and corpses on the seabed 45 meters below. His crew had stumbled onto a Roman cargo ship that sank around 60 BCE, loaded with bronze statues, marble torsos, amphorae, and one shoebox-sized lump of corroded metal that nobody paid much attention to.

That lump sat in the National Archaeological Museum in Athens for nearly fifty years before a British-born historian of science named Derek de Solla Price looked at it closely enough to realise the Greeks had built a hand-cranked analog computer that could predict solar and lunar eclipses using astronomical cycles.

The object is now called the Antikythera mechanism. It contains at least 30 surviving bronze gears, and reconstructions suggest the complete device held closer to 37. Each tooth was cut by hand. Nothing mechanically comparable would exist anywhere on Earth for another 1,400 years.

A box of gears that shouldn’t exist

The mechanism, when it was whole, fit inside a wooden case roughly the size of a hardcover book. A user turned a small crank on the side. Inside, a train of interlocking bronze wheels translated that single rotation into the simultaneous motion of multiple dials on the front and back faces. The front showed the position of the Sun and Moon against the zodiac, the phase of the Moon as a little half-painted ball, and a calendar ring marking the days of the year. The back carried two large spiral dials, one tracking the 19-year Metonic cycle that reconciles lunar months with solar years, the other predicting eclipses across a 223-month Saros cycle.

That last function is the one that still stops people cold. Turn the crank forward a few clicks and the Saros pointer would land on a future date, with engraved Greek glyphs describing whether the eclipse would be solar or lunar, what time of day or night it would occur, and even what colour the Moon might appear during totality. The Greeks had encoded astronomical prediction into a machine you could carry under one arm.

How sponge divers found the wreck

The find was an accident. Around Easter 1900, a crew of Symi divers under Captain Dimitrios Kondos was sailing through the Aegean en route to fishing grounds off North Africa when bad weather forced them to anchor off Antikythera, a rocky speck between Crete and the Peloponnese. A diver named Elias Stadiatis went down in a canvas suit and copper helmet to look for sponges and came up shaking. Kondos went down himself to confirm and surfaced holding the arm of a bronze statue. By November the Greek government and the Hellenic Navy were running one of the first underwater archaeological recoveries in history, with divers hauling up statues, glassware, jewelry, and a few shapeless bronze fragments that crumbled like wet biscuit when they hit the air.

One of those fragments sat in storage at the National Archaeological Museum until May 1902, when Greek politician Spyridon Stais visited the museum to check on the progress of the finds. He noticed a gear wheel embedded in the corrosion of one lump and brought it to the attention of his cousin, the museum director Valerios Stais. Valerios — an archaeologist, not a politician — took up the first serious study of the fragment. Bronze, green with corrosion, with triangular teeth cut at neat angles. Mainstream opinion at the time was that he was looking at a piece of a later instrument that had somehow ended up mixed with the Roman cargo. Greeks of 60 BCE were not supposed to have gears.

Derek Price and the half-century delay

The mechanism mostly gathered dust until 1951, when Derek de Solla Price, then a graduate student at Cambridge working on the history of scientific instruments, began studying photographs and existing publications of the fragments. He made his first direct examination of the bronze in August 1958 during a research trip to Athens funded by the American Philosophical Society, spending ten days handling, measuring, and photographing the pieces. The real breakthrough came in 1971, when Price and the Greek nuclear physicist Charalampos Karakalos of the Demokritos research centre subjected the fragments to gamma- and X-ray imaging that finally revealed the gear trains buried inside the corroded mass.

Price’s 1974 monograph “Gears from the Greeks,” published while he held the Avalon Chair in the History of Science at Yale, laid out the basic architecture of the device and dated it to roughly the second or first century BCE, consistent with the wreck’s amphorae and coins. He suggested the mechanism was a planetarium of sorts. He underestimated it. The full picture only emerged in the 2000s, when teams using high-resolution surface imaging and CT scanning could finally read inscriptions and gear counts that the naked eye and ordinary X-rays had missed.

What the CT scans revealed

In 2005 the Antikythera Mechanism Research Project, a collaboration between researchers at Cardiff, Athens, and Thessaloniki and the technology firms Hewlett-Packard and X-Tek Systems, brought an eight-tonne X-ray tomography machine called Bladerunner to the museum in Athens. The machine, built by X-Tek in Tring and originally designed to inspect jet turbine blades for hairline cracks, fired X-rays through the fragments and produced three-dimensional reconstructions of the interior. Suddenly the engraved instruction manual on the back plates became legible. So did the gear teeth nobody had ever counted properly.

The team led by Tony Freeth and Mike Edmunds counted 223 teeth on one of the back gears. That number is not random. The Saros cycle, which Babylonian astronomers had used for centuries to predict eclipses, runs 223 synodic months — roughly 18 years and 11 days. Other gear ratios matched the Metonic 19-year cycle, the Callippic 76-year cycle, and what looks like the 4-year Olympiad cycle used to schedule the Panhellenic games. The mechanism was not modelling the sky in some abstract way. It was running specific astronomical algorithms that Greek and Babylonian observers had spent centuries refining.

More recent work using 3D X-ray microscopy at University College London has revealed even finer detail, including evidence that the device’s planetary displays once tracked all five planets visible to the naked eye using an arrangement of pin-and-slot gears that mimicked the irregular, retrograde motion the Greeks observed in the night sky.

For most of the last century researchers assumed the calendar ring on the front face marked the 365-day Egyptian solar year. A 2024 analysis by astronomers Graham Woan and Joseph Bayley at the University of Glasgow, using Bayesian statistics and techniques borrowed from LIGO gravitational-wave analysis, concluded the ring most likely had 354 or 355 holes, not 365. That number is the length of a lunar year. The implication, laid out in the team’s announcement of the Horological Journal paper, is that the mechanism tracked the Greek lunar calendar used across much of the Hellenistic world, not the Egyptian civil calendar. The same analysis showed the surviving holes were positioned with remarkable precision, varying on average by only 0.028 millimetres from a perfect circle — accuracy on the scale of a fraction of a human hair, executed in bronze, by hand, in a workshop somewhere in the eastern Mediterranean.

It is a small correction with large consequences. A lunar calendar reading means the device was tuned to the religious and civic life of a specific Greek polity, not to an imported Egyptian timekeeping system. Somebody in the eastern Mediterranean wanted to know exactly when the next festival fell, when to expect the next eclipse, and where the Moon would sit in the zodiac on the night their daughter got married.

Who built it, and where

Nobody knows for certain. The engravings are in koine Greek, the common dialect of the Hellenistic world. The astronomical conventions point to Rhodes or possibly Syracuse. Cicero, writing a few decades after the shipwreck, describes a similar bronze device built by Archimedes that showed the motions of the Sun, Moon, and five planets, and another made by the Stoic philosopher Posidonius on Rhodes. The Antikythera mechanism is not Archimedes’ machine, but it sits in the same lineage.

The Roman ship was probably ferrying looted Greek treasure to Italy. Generals like Sulla had been stripping Greek cities of their art and instruments for decades. The mechanism may have been a custom commission for a wealthy Roman buyer who wanted a piece of Greek scientific glamour to display in his villa. Instead it went to the bottom of the Aegean and got buried under two thousand years of sediment.

The 1,400-year silence

What makes the object genuinely strange is what happens after it sinks. The technical knowledge required to design and cut its gears — pin-and-slot followers, accurate tooth-count ratios encoding astronomical cycles, an internal train of more than thirty meshed wheels — simply disappears from the historical record for over a millennium. The next devices that approach this level of mechanical complexity are the astrolabes and geared calendars produced in the Islamic world from roughly the 11th century onward, and the astronomical clocks built in medieval European cathedrals from the 14th century. The technology was not lost in some single catastrophe. It was slowly forgotten because the world that valued it stopped existing.

There is no evidence that the Antikythera mechanism was unique in its own time. The references in Cicero, the existence of multiple gear styles within the device itself, and the polished quality of the engraving all suggest a workshop tradition with multiple craftsmen and earlier prototypes. But those workshops, and whatever other machines they produced, vanished. The Antikythera fragments are the only direct material trace that survives.

The reconstruction that finally worked

In March 2021 a UCL team led by Tony Freeth published a complete mechanical reconstruction in Scientific Reports of the front cosmos display that matched every surviving fragment and every legible inscription. Their model uses a complex stack of nested rings driven by pin-and-slot followers to reproduce the irregular motions of Mercury, Venus, Mars, Jupiter, and Saturn as the ancient Greeks saw them — the loops and reversals that puzzled astronomers until Kepler explained them with elliptical orbits seventeen centuries later. The resulting reconstruction reconciled the planetary inscriptions with the surviving gears for the first time, and subsequent coverage in The Jerusalem Post described how the work continues to upend conventional timelines of ancient technology.

Curious readers can see working physical models in several museums, including the original Athens fragments housed in the Bronze Collection of the National Archaeological Museum. For a sense of how far mechanical computing has come since, it helps to compare the gear-train logic of the Antikythera device to the binary switching that underpins how a modern processor handles calculations at the silicon level. Both are doing the same fundamental thing — translating a single input, a crank turn or an electrical pulse, into a controlled cascade of state changes that arrive at an answer about the world. The Antikythera mechanism stands a long way from the silicon of a modern chip, but it sits on the same line, two millennia further back than most popular histories of computing bother to look.

What it sounds like to use

Working replicas exist, and the experience of turning the crank is oddly intimate. A faint clicking, the soft brass-on-brass slip of well-cut teeth, the slow rotation of pointers across painted dials. One full turn of the input shaft moves the Sun pointer forward by roughly one day. Spin it through a year and the little half-black-half-white lunar phase ball rotates twelve and a third times. Spin it through nineteen years and the long spiral pointer on the back returns almost exactly to its starting position, having drawn out 235 lunar months in a perfect Metonic cycle.

The bronze, when freshly polished, would have been the colour of warm honey. The case sat on a table. Somebody in a Hellenistic city, probably wearing wool and standing in afternoon light filtered through a colonnade, turned that crank and watched the Moon move. They knew, weeks or years in advance, when the sky would go dark. The machine that told them this is currently in three large fragments and dozens of smaller ones, stored in climate-controlled cases on the ground floor of the National Archaeological Museum in Athens, still slowly being read.

Subscribe to our newsletter!

Our latest tutorials delivered straight to your inbox

Make Tech Easier Editorial Team Avatar

Read next

In November 1988, an unmanned Soviet space shuttle called Buran flew a full orbital mission and landed itself in a blizzard at Baikonur without a single human input, and three years later the country that built it no longer existed.
When Boeing 747-400 pilots needed to update their navigation database as late as 2020, a technician would walk onto the flight deck with eight 3.5-inch floppy disks and feed them one at a time into a slot beside the captain’s seat, because recertifying anything newer than the 1989 avionics would have cost hundreds of millions of dollars.
When the U.S. military finally deployed frequency-hopping spread spectrum on Navy ships during the 1962 Cuban Missile Crisis blockade, the patent that described the technique had expired three years earlier, and its inventor was watching the news in a Los Angeles bungalow without any idea her idea was at sea.
Suzanne Simard sealed paper birch and Douglas fir seedlings inside plastic bags, fed them carbon-14 and carbon-13 dioxide, and nine days later found carbon had crossed between species through fungal threads in the British Columbia soil beneath her boots
In the small hours of 2 September 1859, a telegraph operator in Portland, Maine disconnected his batteries because they were throwing sparks, and then discovered he could still send a clean message to Boston using nothing but the current the aurora was pushing through the wire above his head.
In 1992, a container ship leaving Hong Kong lost 28,800 plastic bath toys overboard in the North Pacific, and oceanographer Curtis Ebbesmeyer spent the next two decades tracking yellow ducks and blue turtles as they washed up in Alaska, Maine, and eventually the coast of Scotland, quietly rewriting the textbook map of ocean currents.
In 1942, Hollywood actress Hedy Lamarr and avant-garde composer George Antheil received US Patent 2,292,387 for a frequency-hopping radio system synchronised by a perforated paper roll borrowed from a player piano, a technique the Navy filed away as unworkable and which now underpins every Wi-Fi router, Bluetooth earbud, and GPS receiver on Earth.
In 1979, a Sony engineer named Nobutoshi Kihara built the first Walkman prototype in four days because his boss Masaru Ibuka wanted to listen to opera on long flights, and the team launched it with no advertising budget, no headphone jack standard, and an internal forecast of 5,000 units a month that the device beat in its first fortnight.