How Roman aqueducts were built: surveying a slope of inches per mile, tunnels, siphons and arches, who did the work, and why the water still runs

Architecture · 4 September 2026

A Roman aqueduct was a channel that carried water from a spring or river to a city by gravity alone, falling a few feet in every mile, and the arches everyone pictures were the smallest part of it: of the 260 miles of Rome's eleven aqueducts, about 30 ran on arches, and the rest in a covered masonry channel cut into hillsides or buried a few feet down, with tunnels through ridges and, where a valley was too deep for arches, a siphon of lead pipe that took the water down one side and up the other. They were built between 312 BC and AD 226 for Rome, and for hundreds of cities of the empire from Segovia to Carthage, by surveyors with a water level and a sighting cross, by soldiers, contractors' gangs and slaves with picks, by masons who cast concrete and cut stone, and they were paid for by conquest, by emperors and by rich men who wanted their names on them. Below: how the source was chosen, the survey, the channel, the tunnels and siphons, the arches and how they were built, the mortar and the tanks, who built them and how long it took, Rome's system and its manager Frontinus, and the aqueducts that still run, with the Pont du Gard and the Aqueduct of Segovia in the Buildings by style collection and the material in Roman roads and concrete.

Finding the water

Vitruvius, writing about 25 BC, tells the engineer how to find a spring: look at dawn for mist rising from the ground, dig where rushes and willows grow, bury a bronze bowl overnight and see if it sweats, and test the water by the health of the people who drink it and by boiling it in a bronze pot to see if it leaves sand or sediment. Rome's aqueducts took water from springs in the limestone hills of the Anio valley 25 to 40 miles east, and from the river Anio itself for the lower-grade supply, and the source had to be higher than the city by enough to give the channel its fall over the whole distance: the Aqua Claudia's springs were at 320 m and the channel delivered at 67 m in Rome after 69 km, an average fall of 3.6 m per km, about 1 in 280. The higher the source, the higher the water could be delivered in the city, and the best water, cold and clear, went to the drinking fountains and the emperor's houses; the muddy Anio Novus went to the baths and the gardens. Roman aqueduct.

The survey

The channel had to fall the whole way, never rise, and not fall too fast, which would scour it, or too slowly, which would silt it; Vitruvius asks for a minimum of 1 in 200 and the surveyed aqueducts vary from 1 in 100 to, at Nîmes, 1 in 3,000 for a stretch, a fall of 34 cm in a kilometre, which the surveyors held over 50 km. Their instruments were the chorobates, a wooden bench 20 feet long with a water trough in the top and plumb lines at the ends, levelled and sighted along; the groma, a cross of sighting arms on a staff with plumb bobs, for right angles and straight lines; the dioptra, a sighting disc on a stand with a water level, described by Hero of Alexandria; and measuring rods and cords. The route was chosen to follow the contours around the heads of valleys rather than cross them, which is why aqueducts wind, the 50 km of the Nîmes aqueduct covering a straight distance of 20; the surveyors set out the line with stakes and marked the levels, and at Nîmes the fall over the last 6 km was held to 7 mm per 100 m. Mistakes were made: an inscription from Saldae in Algeria records the engineer Nonius Datus, sent back to find that the two ends of his tunnel, dug by the contractor's men from both sides, had each passed the other in the rock.

The channel

• Specus. The channel, about 0.5 to 1 m wide and 1.5 to 2.5 m high, of concrete or masonry with a vaulted or slab roof, lined with waterproof mortar and running about two-thirds full; a man could walk it for cleaning, and manholes every 70 m or so gave access.
• Underground. Most aqueducts were cut as a trench, built, and covered, or tunnelled in the hillside a few metres down, which kept the water cool, hid it from enemies and thieves and protected it from frost; the arches were used only where the channel had to keep its height across a valley or the last approach to a city on a plain, as at Rome, where the Claudia and Anio Novus ran on arches for the last 10 km to reach the high ground.
• Tunnels. Cut through ridges by gangs working from vertical shafts sunk every 30 to 60 m along the line and digging toward one another at the bottom, the shafts serving to lift the spoil and to keep the line straight and level; the tunnel of the Anio Novus at Subiaco and the 1 km tunnel of the Nîmes aqueduct at Sernhac were cut with picks in limestone, at perhaps a metre a day per face.
• Bridges. Where a stream crossed the line, a small bridge; where a valley was less than about 50 m deep, an arcade.
• Siphons. Where a valley was deeper, the water was collected in a tank at the top of one side, run down a battery of lead pipes, across the valley floor on a low bridge and up the other side to a tank a little lower than the first, the pressure of the descent pushing it up; the aqueducts of Lyon had nine siphons, one 120 m deep with 2.6 km of pipe, and the Pergamon siphon of the 2nd century BC was 190 m deep. Lead pipe was made from sheet, rolled and soldered, in standard sizes; the Romans knew lead poisoned water and Vitruvius preferred clay, but the siphons needed lead.

The arches

An arcade was a row of piers of concrete faced with stone or brick, or of cut stone, carrying arches on which the channel ran, and its height, sometimes tiered, kept the channel level across a valley. The arches were built on wooden centering, a timber frame the shape of the arch's underside set on ledges or on the piers, on which the voussoirs, the wedge stones, were laid from both sides up to the keystone, after which the frame was struck and moved to the next arch; the ledges and the holes for the scaffolding beams are still on the Pont du Gard, and the stones there, of up to 6 tons, were lifted by cranes with lewis holes and treadwheels. The Pont du Gard, built about AD 40–60 to carry the Nîmes aqueduct across the Gardon, is 49 m high in three tiers, 275 m long, of stone cut so precisely that it was laid without mortar, and it carried the channel with a fall of 2.5 cm over its length; the Aqueduct of Segovia, of about AD 100, is 28 m high and 813 m long on 167 arches of unmortared granite, and carried water into the town until the 20th century. Rome's arcades were concrete and brick, cheaper and faster; the Aqua Claudia's arches, 27 m high, ran for 10 km across the Campagna and are the ruins in every painting of the Roman countryside.

Mortar, tanks and the city

• Waterproofing. The channel was lined with opus signinum, mortar mixed with crushed tile or brick, which the Romans knew set hard under water, laid in several coats and rounded at the corners; the lime deposit (sinter) that the hard water left inside had to be chipped out and is now used to date and read the flow.
• Settling tanks. Piscinae limariae along the line, where the water slowed and dropped its sand, cleaned regularly.
• The castellum. At the city the water entered a distribution tank, the castellum divisorium, from which it was divided by pipes into three parts, in Vitruvius's plan, for the fountains and basins, the baths and the private houses, in that order, so that in a shortage the public supply failed last; the castellum at Nîmes survives, a round basin with ten outlet pipes and three sluices, and the one at Pompeii with its three channels.
• Pipes. Lead pipes, fistulae, stamped with the name of the owner and the plumber, carried water from the castellum to the fountains, baths and licensed private users, through bronze nozzles (calices) of a fixed size that measured the grant; Frontinus lists 25 sizes.
• Overflow. The water never stopped: the fountains ran day and night, and the overflow flushed the sewers, the Cloaca Maxima and its branches, into the Tiber, which is why Rome, with a million people, was cleaner than any city until the 19th century.

Who built them and how long it took

Rome's first, the Aqua Appia of 312 BC, was built by the censor Appius Claudius, who also built the road, and the second, the Anio Vetus of 272, from the spoils of the war with Pyrrhus; the Marcia of 144 BC cost 180 million sesterces from the spoils of Carthage and Corinth, and Agrippa built two for Augustus with his own money and his own gang of 240 slaves, which he left to the emperor. The Claudia and Anio Novus, begun by Caligula in 38 and finished by Claudius in 52, cost 350 million sesterces and took 14 years, with a workforce in the tens of thousands, mostly contractors' men and slaves, with the army's surveyors; in the provinces the legions built aqueducts as they built roads, and cities paid for their own by subscription, by a benefactor's gift or by a loan from the emperor, the Nîmes aqueduct being a gift of the city's elite, with no emperor's name on it. Repairs were constant: Frontinus, the water commissioner of Rome in 97, found the channels riddled with illegal taps by landowners and the water carriers' fraud, the sinter choking the Marcia, and ordered the arcades of the Claudia repaired, which had been leaking for years; his book "On the Aqueducts of Rome" is the manual, and gives the figures.

Rome's eleven and the ones still running

• Rome. Eleven aqueducts by 226, about 800 km of channel including branches, delivering perhaps a million cubic metres a day, or up to 1,000 litres per person, more than modern Rome; the Goths cut them in the siege of 537 and only the Aqua Virgo of Agrippa, 19 BC, mostly underground, ran on through the Middle Ages and runs today, feeding the Trevi Fountain and the fountains of the Piazza Navona and the Spanish Steps. The Aqua Traiana of 109 was restored by Paul V in 1612 as the Acqua Paola and runs.
• Segovia. Supplied the town until 1973 and carries water still, in a modern pipe on the Roman arches.
• Nîmes and the Pont du Gard. Ran from about AD 50 to the 6th century, choked with sinter; the bridge carried a road from the 18th century and is a World Heritage Site with a million visitors a year.
• Still in use elsewhere. The aqueduct of Valens in Constantinople, of 373, fed the city until the 20th century; the Roman channels at Évora in Portugal, at Carthage (restored in the 19th century) and at Caesarea are in part in use; the Roman springs and lines in many cities of France and Spain are under the modern pipes.
• The longest. The Zaghouan aqueduct to Carthage, 132 km; the Gadara aqueduct in Jordan, 170 km, with a 94 km tunnel; the Anio Novus, 87 km, the longest at Rome.

Versions and theories

• Were the aqueducts built by slaves? In part: Agrippa's gang of 240 was slaves, and contractors used slaves and free labourers together, as on all Roman building; the surveying and engineering was by free men and army specialists, and the provincial aqueducts were often built by soldiers.
• Lead poisoning brought down Rome. The Romans used lead pipes and lead-lined vessels, and their bones carry more lead than ours; but the water ran continuously and the hard water coated the pipes with lime, so the pipes gave little, and the empire had other problems. The theory is from 1983 and does not hold.
• The Pont du Gard was built without mortar. The stone tiers, yes, set dry with the stones cut to fit; the channel on top is concrete lined with waterproof mortar, and some stones were fixed with iron clamps.
• The Romans did not know the siphon. They knew it from the Greeks and used it whenever a valley was too deep for arches; the myth arose because the Roman texts prefer arches and the pipes were melted for their lead.
• The arches are the aqueduct. The arches are the bridges; 80 to 90 per cent of any Roman aqueduct is underground or at ground level and invisible, which is why the ruins in the Campagna give a false picture.
• The Segovia aqueduct was built by the Devil in a night. The local legend, with a girl who sold her soul for water and was saved when the last stone was not set before dawn; the missing stone in the legend is a real gap that held a bronze inscription, restored in letters of the 1990s.

The roads that went with them are in Roman roads and concrete, and the vaults the same builders raised in who built the Pantheon.

Open the sections: buildings by style, ancient temples of the world, all articles.