It is a lump of green corrosion about the size of a thick book.
For months it sits in a crate in Athens while everybody looks at the statues instead. Then the wood dries out, the lump splits along a seam, and there is a wheel inside it.
A bronze wheel. Teeth cut all the way around the rim, dozens of them, each one shaped by hand. Behind it, more wheels, pressed flat into one another like the pages of a soaked book.
It came off the seabed with marble limbs and glassware and a bronze youth, all of it roughly two thousand years old. And nothing remotely like that wheel turns up anywhere on Earth for the next thousand years.
What the divers found
They were not looking for it. They were looking for shelter.
A crew of sponge divers from the island of Symi was working its way home from the grounds off North Africa when the weather turned. They put in at Antikythera, a bare scrap of rock in the channel between Crete and the Peloponnese. Accounts differ on the month. The spring of 1900 is the date usually given.
While they waited out the wind, the captain, Dimitrios Kontos, sent a man down to see whether the sponge was worth taking. Elias Stadiatis went over the side in canvas and a copper helmet, to something near fifty meters.
He came up talking about corpses. A heap of dead naked people on the bottom, he said, and horses among them.
Kontos went down himself. He came back holding an arm. It was bronze.
What they had blundered into was a cargo ship that had gone down loaded with Greek art, most likely bound for Italy, at some point in the first century BC. The divers reported it in Athens. The Greek navy sent a vessel, and the salvage ran through the winter and into 1901.
The cost was real. Working that deep in that equipment, with nobody yet understanding what pressure does to blood, one diver died and others were left paralyzed. Statues came up with one side eaten smooth by the sea and the buried side still crisp. Coins came up. Couches, jars, a bronze youth with his arm raised.
And one unremarkable lump of encrusted metal that nobody put at the top of the list.
The wheel in the crate
The usual account credits the archaeologist Valerios Stais, at the National Archaeological Museum in Athens, with noticing the gear on 17 May 1902.
The reaction was not celebration. It was suspicion. A toothed wheel implied a machine, and no machine of that kind was supposed to exist in the ancient world. The obvious explanation was that something later had fallen into the site. A medieval instrument, dropped from a passing boat, settling among Roman cargo.
That explanation did not hold. The lump had come up inside the wreck material, and the wreck dated to somewhere around the middle of the first century BC.
So the wheel was Greek. Which meant somebody had to work out what it was for.
Turn the handle and the sky moves
Here is the thing at full strength.
The mechanism was a wooden-framed case you could carry under one arm, filled with interlocking bronze gears. On the side was a crank. You turned it, and the pointers on the faces moved, and what they showed you was the sky on any date you chose.
The front carried a ring of the zodiac and a ring of the Egyptian calendar. Pointers tracked the sun and the moon through them. A small ball, dark on one side and light on the other, rotated to show the phase.
The back is where it gets serious. Two large spiral dials, cut in several turns. The upper one runs the Metonic cycle: nineteen years, two hundred and thirty-five lunar months, the interval in which the moon's phases come back into step with the solar year. The lower one runs the saros, the two hundred and twenty-three month cycle after which eclipses repeat. That dial did not merely count. It named which months carried an eclipse, and researchers reading the inscriptions have recovered predicted times and characteristics.
Smaller subsidiary dials handled longer corrections, and one four-year dial appears to have tracked the cycle of athletic games.
Now the detail that stops people.
The moon does not move at a constant speed. It runs fast at one part of its orbit and slow at another. The mechanism reproduces that. Two gears are stacked face to face on slightly offset axes, with a pin on one riding in a slot cut in the other. As they turn, the pin is forced to sweep unevenly, and the output speeds up and slows down across the month.
That is epicyclic gearing, built to model an astronomical irregularity, in bronze, before the Roman empire existed.
Eighty-two fragments are known. Around thirty gears survive. Every serious reconstruction requires more than that, gone to the sea.
How it was read
Most of what we know came from looking through the corrosion rather than at it.
Derek de Solla Price, a historian of science at Yale, spent decades on the problem. He wrote it up for Scientific American in 1959 under the title "An Ancient Greek Computer." In 1974 he published a full monograph, "Gears from the Greeks," working with the Greek radiographer Charalampos Karakalos, whose gamma-ray images allowed teeth to be counted beneath the crust for the first time.
Price got a great deal right and some things wrong. Michael Wright, then a curator at the Science Museum in London, pushed further with linear tomography and built working models.
Then in 2005 a purpose-built X-ray CT scanner weighing several tonnes was shipped to Athens to image the fragments. The results appeared in Nature in 2006, from a team including Tony Freeth and Mike Edmunds. A further Nature paper in 2008 reported the games dial and month names that some researchers read as Corinthian, which would point the workshop west, toward Syracuse. In 2021 Freeth and colleagues published a proposed reconstruction of the front display showing the planets.
Every round of better imaging has made the device more sophisticated, not less.
The conventional account, and it is a good one
Mainstream history of science does not treat this as an anachronism, and the argument deserves stating properly.
Cicero, writing within a lifetime of the wreck, describes bronze spheres that modeled the heavens. One he attributes to Archimedes, carried off from Syracuse by the general Marcellus when the city fell in 212 BC. Another he credits to Posidonius, in his own time. The texts, on this reading, tell us such instruments existed and that educated Romans knew of them.
The standard account adds two more points. Bronze was valuable and endlessly recycled, so survival of any bronze device requires an accident like a shipwreck. And the astronomy inside the machine was not new. Babylonian scholars had worked out the saros and the moon's variable speed centuries earlier. The theory was on the shelf. What the Greek workshop added was a way to turn arithmetic into brass.
By that account the mechanism is not a miracle. It is the one survivor of a craft we know existed and whose other products went into the melt.
The part that will not go away
It is not a first attempt.
Look at what the object actually required. Tooth counts chosen so the ratios come out as astronomical periods. Gear trains routed around each other inside a case the size of a book. A pin-and-slot coupling invented to reproduce a specific irregularity in the moon. Inscriptions covering the plates, apparently instructing a user.
Nobody reaches that on the first try. It implies a workshop, drawings, failures, apprentices and earlier versions.
None of that has been found. Not a spare gear. Not a blank. Not a tool unambiguously made for cutting teeth of that size.
Then there is the gap. The next geared calendrical instrument that survives in the hand is Islamic: a geared astrolabe signed by Muhammad ibn Abi Bakr of Isfahan, dated to 1221 or 1222, now in London. Al-Biruni had described the principle around two centuries before that. Between the Antikythera wreck and those objects lies more than a thousand years of Mediterranean and Near Eastern archaeology with no comparable machine in it.
Preservation may account for that. It is also true that a millennium of digging has yet to produce one stray toothed bronze wheel from the tradition the mechanism is supposed to belong to.
Even its age is unsettled. Price put its construction near 82 BC. Later work pushed it earlier, into the second century BC. In 2014 Christian Carman and James Evans argued from the eclipse dial that its start date was 205 BC, which would place it within a decade of the death of Archimedes.
What is left is the metal
Eighty-two fragments in a case in Athens. Thirty wheels. Teeth a millimeter or two across, cut by somebody whose name nobody knows.
Either it is exactly what the standard account says. The lucky survivor of a real and ordinary craft, whose brothers were melted for coin and cookware, and the silence around it is only the silence of recycling.
Or something happened in one Greek workshop that the rest of the record has no place for, and the thousand years of nothing afterward mean what they appear to mean.
The bronze has not said which.