How a Roman Bath Was Heated

Vitruvius’s instructions for a hot room begin at the bottom. Pave the ground with tiles a foot and a half square, he tells the builder, sloping towards the furnace, “so that a ball thrown in cannot stay inside but comes back of its own accord to the praefurnium”; then “the flame will wander more easily under the suspended floor”. On the pavement go stacks of small square bricks, bessales, two feet high, spaced so that tiles two feet square can bridge from stack to stack, and bedded in clay worked with hair, because lime mortar would burn. The big tiles carry the floor (Vitruvius V.10.2).

The ball is a builder’s test, and it says something about the gas. A floor that falls towards the fire rises away from it, so the hot smoke, which climbs wherever it can, is led uphill towards the far wall and the flues. Vitruvius’s flame is a writer’s word. The fire stayed in the mouth of the furnace, and what wandered under the floor was smoke and hot air, pulled along by a weak, steady draught. The art of the system lay in making that current give up its heat to the floor and walls before it got out.

This underside is the part of a Roman bath that survives. At Kourion, Amathus and Salamis a visitor finds a field of knee-high brick stacks on a tiled floor, sometimes with a slab of the floor they carried still lying across their heads. This report puts the rooms back, lights the fire, and follows the heat.

No excavated bath survives well enough to rebuild on its own, so this one is assembled. It belongs to what Daniel Krencker in 1929 called the row type: rooms in a line, with the bather walking out to the hottest room and back the same way, through the same door (Nielsen 1990; de Vries & Lain 2006, p. 359). Baths of this size were everywhere. Rome’s 4th-century regionary catalogues list 856 balneae beside ten or eleven great thermae (Yegül 2013, p. 3).

The proportions come from two small baths dug and measured in full: the fort bath at Bearsden on the Antonine Wall, of the mid-2nd century AD, 28 m along its axis with a hot room of 3.5 by 3.1 m (Breeze 2016, pp. 48, 61–63), and the legionary bath at el-Lejjun in Jordan, of about AD 300, 23 by 9.7 m overall (de Vries & Lain 2006, pp. 355–359). Ours is about 25 m long and 3.8 m wide inside.

The Cypriot details come from C. A. Harvey’s study of how baths were built on the island (Harvey 2024) and from the two Cypriot baths he counts as published in detail: the bath outside the Sanctuary of Apollo Hylates at Kourion, dated by an inscription to about AD 101/2 (Scranton 1967, pp. 56–61), and the bath wing of the House of Orpheus at Nea Paphos, of about AD 200. So the walls are limestone in lime mortar, as Cypriot bath walls were.

The one door is in the south wall of the changing room, the apodyterium. Bathers undressed there and left their clothes in wall niches, watched by their own slave or a paid attendant. The niches follow el-Lejjun, where seventeen lined the changing court above a plastered bench (de Vries & Lain 2006, p. 356).

The cold room, the frigidarium, has a plunge sunk in its floor against the north wall, a tub for a quick immersion; swimming pools belong to the great thermae. Bearsden’s was 3 m square and 0.70 m deep (Breeze 2016, p. 54). The next door is narrower. At el-Lejjun the door from the cold room to the warm room was 0.66 m wide, the narrowest in the building, to keep the heat in (de Vries & Lain 2006, p. 357); ours is 70 cm.

The warm room, the tepidarium, is heated second-hand. Beyond it the hot room, the caldarium, has the hot tub, the alveus, across its east end over the furnace, and in an apse on its south side the labrum, a raised basin of cooler water, under the window so that the people round it would not shade it (Vitruvius V.10.4). The usual round went out to the hot room for sweating, a soak and a splash at the labrum, then back to the cold plunge, with oil and a strigil to scrape it off.

Bathing was an afternoon habit, “mainly from midday to evening” in Vitruvius (V.10.1). At the mining settlement of Vipasca in Portugal the lessee had to heat the bath every day, for women from daybreak to the seventh hour and for men from the eighth hour to the second hour of the night, which meant lamps (Lex metalli Vipascensis, CIL II 5181). A bath with one suite of rooms was shared by the clock.

The furnace room, the praefurnium, is a lean-to against the east end, its floor sunk 95 cm below the ground so that the fire sits level with the void under the hot rooms. At Apollo Hylates the furnace room was sunk about 1.60 m below the caldarium floor, with the fire entering the bath through a narrowing tunnel between short brick walls (Scranton 1967, p. 58).

Our furnace is a brick channel 55 cm wide and 70 cm high under an arch, through the 1.1 m wall at the hot end; Bearsden’s was 550 mm wide (Breeze 2016, pp. 64–65). There is no grate. Roman furnaces burned on the floor of the channel, with the air drawn over the fuel, and a fire fed that way burns slowly and can be left for long stretches (Rook 1992, via Goulianou 2020, p. 35).

The fuel was wood. Charcoal from the furnace of a Romano-British villa proved to be mostly oak, with some poplar or willow, and its reflectance under the microscope, which rises with the heat it was burnt at, puts the fire at 330–410 °C (McParland et al. 2009). We know of no analysis of charcoal or ash from a Cypriot furnace. Troodos pine, olive prunings, carob wood and the pressed waste of olives were all to hand; which fed Cypriot baths is our guess, and the model burns plain wood.

The stoker, the fornacator, was usually a slave, and when a bath changed owners he went with it, as part of its equipment (Meusel 1960, cited by DeLaine 1988). In a reconstructed hypocaust room at the Saalburg fort in Germany, Hüser kept a fire in for five days on three stokings a day, at an average of 1.6 kg of wood an hour (Hüser 1979); the team that built a bath at Sardis in 1998 stoked every two hours, day and night (Yegül & Couch 2003). At Vipasca the lessee had to keep a stock of firewood for a set number of days and was fined 100 sesterces for every sale of it.

Under the warm and hot rooms the model stands 80 pilae on 60 cm centres, so that each big floor tile, a bipedalis 59 cm square, rests on the corners of four stacks. A stack is a base tile, square bricks 20 cm across in 6 cm courses, and a broad cap: about 700 bricks in all. They stand 0.60 m high at the far end, Vitruvius’s two feet, and a little more by the furnace, where his slope has taken the floor down 6 cm.

Cypriot pilae, on the evidence Harvey collected, are all fired brick. Square stacks stand at Apollo Hylates, of tiles about 21 cm square, at Amathus and in the Baths of Eustolios at Kourion; round ones at Salamis and in the House of Orpheus at Nea Paphos (Harvey 2024; Scranton 1967, p. 58). The one Cypriot pila height we know of is from the House of Orpheus, where round bricks 20 cm across made stacks that “once stood 60 cm high” (Harvey 2024, §26).

The stacks are tall for a reason. Gas in a hypocaust settles in layers, hottest against the underside of the floor, and a deep void keeps the heat up against the tiles meant to take it (Rook 1978, via Maréchal 2020, p. 49). What moves it is the draught of the flues: a column of warm gas in the walls is lighter than the air outside, and the difference pulls gas out of the furnace, across the hypocaust and up. For a flue five metres high at 100 °C, with 15 °C outside, a rough calculation gives a pull of about 13 pascals. With too few outlets the fire smoulders and smoke backs out of the stoke-hole; with too many, it burns fast and eats wood (Goulianou 2020, pp. 35–36, 43).

Our model follows the gas along its path. It leaves the fire at about 330 °C, close to the charcoal’s 330–410 °C, gives heat to the tub and the floor, and averages about 260 °C under the caldarium and 180 °C under the tepidarium. Those are averages over each stretch in a lumped model, and they run well above the 90–125 °C reported under the floors of the Indirizzo baths at Catania (Energy and Buildings 138, 2017). Read them as an upper bound: our model of the bath drawn here gives the order of things along the path, set beside the measurements.

On the pilae lies the suspensura, the hanging floor: 105 bipedales, a bed of opus signinum (lime with crushed brick) 15 cm thick, and the finish, marble in the hot room and mosaic in the warm. Maréchal gives 15–20 cm for the concrete (Maréchal 2020, p. 46); at Apollo Hylates a rubble bed about 15 cm thick lay over two courses of tiles (Scranton 1967, p. 58). Ours is 23 cm thick, and over the 30 m² of the two heated rooms it weighs about 14 tonnes.

That weight is the system’s store. The gas heats the floor from below, the floor heats the room from above, and it goes on doing so after the fire dies down. Thickness sets the balance: at Xanten, Grassmann found that a floor up to 60 cm thick brought the surface down from 27.5 °C to 22 °C (Goulianou 2020, p. 44). A thin floor gives a hot surface quickly and loses it quickly; a thick one takes days to charge and keeps what it takes.

The heat was never even. When Hüser measured the Saalburg room in 1979 the floor beside the furnace was at 38 °C against an average of 18 °C, and the far corner at 9 °C (DeLaine 1988). Ancient stories of scalding floors belong by the furnace. Pliny tells how the slaves of Larcius Macedo beat their master in his bath and threw him onto the “scalding floor” to find out whether he was still alive (Pliny, Letters 3.14.2).

In our model the caldarium floor settles at about 56 °C by the tub and 47 °C at the far end: warmer than floors measured in rebuilt baths, 38 °C by the furnace at the Saalburg and 20–50 °C at Carnuntum in Austria, and more even, since the Saalburg spread from 38 °C to 9 °C is wider than a lumped model can give. Either way it is a floor to cross in sandals.

In about AD 64 Seneca listed things that had appeared within living memory: windows of translucent panes, the suspended floors of baths, “and pipes set into the walls, through which the heat is spread so as to warm the lowest and the highest parts equally” (Seneca, Letters 90.25). He was wrong about the floors, which go back to the late 2nd century BC in Campania, and right enough about the pipes. The first hollow walls, at Pompeii, were flat tiles with a boss at each corner, tegulae mammatae, nailed over a slot too narrow for the gas to rise freely. From the 1st century AD builders used tubuli, box tiles open at both ends, set side by side up the wall and plastered over (Maréchal 2020, p. 47). Vitruvius does not mention wall heating at all (Maréchal 2020, p. 48).

Whether the walls heated the room or only kept its heat in was argued for forty years. In 1953 Kretzschmer fired the Saalburg room on charcoal and concluded that the hollow walls merely insulated (Kretzschmer 1953). Hüser’s re-run in 1979, burning wood and using thermal imaging, found the walls on average as warm as the floor (DeLaine 1988), and Grassmann’s measurements at Xanten in 1993–94 showed the box tiles heating actively (Maréchal 2020, pp. 48–49). Grassmann went further, arguing that the gas turned over inside the wall, two columns rising hot beside two sinking cooler, linked by the vents in their sides. Schiebold rejected that in 2006, and the point is still disputed; the model shows the flow as mostly upward, with some mixing sideways.

The Sardis bath taught a practical lesson. With every column opening into one flue along the wall head, the gas ran straight up the columns nearest the furnace and the far ones stayed cold. With only every fifth row connected, the heat spread evenly and the fire burned slowly and economically (Yegül & Couch 2003, via Maréchal 2020, p. 49).

Cypriot builders used two systems, and the model offers both. The large baths on the Kourion acropolis, the gymnasium baths at Salamis and the late bath at Agios Georgios used box tiles of a local design, narrowed at one end to socket into the next and grooved on the short sides for an iron nail, 32–34 cm high, 24–28 cm wide and 17–19 cm deep (Harvey 2024, §38). The default model lines its hot walls with about 380 of them. The bath at Apollo Hylates had something else: terracotta spacer pegs 26 cm long, grooved and set in sockets on a grid about 0.50 m apart, holding a shell of thin tiles over a cavity 10–15 cm deep, mortared and faced with marble (Scranton 1967, pp. 59–60). Kourion’s are the only rectangular pins known, and among the longest (Farrington & Coulton 1990, p. 65). The same bath had flues cut into its walls, and its suspended floor stopped about 15 cm short of them, so that gas could rise up the wall face. Spacer pins turn up in the House of Orpheus and the Amathus agora baths too.

The model can show what the walls are worth by taking them away. With the floor alone and a single chimney, our caldarium air settles at about 25 °C, against 35 °C with the box tiles. The Sardis team reached the same answer from the other side: without wall heating, their floor would have had to run at about 62 °C to give the room they wanted, too hot to stand on (Yegül & Couch 2003, via Goulianou 2020, p. 43). Heated walls are what let a hot room have a window and stay hot (Maréchal 2020, p. 50).

How the gas left the building is among the least settled parts of the system. Maréchal describes chimneys in the corners of rooms, drawing the gas through the walls to the roof (Maréchal 2020, p. 47); Rook held that Roman baths “did not have chimneys as such”, only vents and short ducts working as flues (Rook 1992, via Goulianou 2020, p. 35). At Apollo Hylates the wall flues became enclosed in the masonry about 1.10 m above the floor and carried the draught upward (Scranton 1967, pp. 58–61); where they came out has not survived.

The model gives the gas a row of terracotta spouts through the wall heads, 14 by 12 cm, about every 1.2 m, just under the springing of the vault: a reconstruction, chosen as the simplest arrangement that fits a vaulted room with lined walls. In our model the gas leaves at about 70 °C. We know of no measurement at the outlets of a Roman bath. While the system was sound, smoke never entered the rooms; a crack in the floor or the lining let it in, with its carbon monoxide.

Vitruvius sets three bronze vessels over the furnace, hot, tepid and cold, arranged so that as much hot water as runs from the tepid vessel into the hot one is made up from the cold into the tepid; “and the testudines of the tubs”, he adds, “are heated from the same furnace” (V.10.1). The testudo, the tortoise, was a bronze half-cylinder over the furnace channel, open to the floor of the tub above. Cold water from the bottom of the tub sank into it, warmed and rose again, so the tub kept itself hot by circulation (Maréchal 2020, p. 46).

Excavated baths rarely show three vessels. They show one boiler, two at most, set over the fire on brick walls and iron bars and cased in brickwork, with tepid water made by mixing (Maréchal 2020, p. 46). Bronze was worth reusing and every boiler was salvaged, so their places are read from the masonry; at Bearsden the wide cheeks of the furnace probably carried the boiler (Breeze 2016, pp. 64–65).

In the model a bronze half-cylinder 0.68 m long hangs in the channel under the tub, and a bronze drum 0.72 m across stands on a masonry block over the channel’s mouth, where the gas is hottest. Lead pipes run from a tank to the boiler and the plunge, and from the boiler to the tub, which holds about 2,000 litres and settles at about 40 °C in our model. Measured tubs run from 30 °C (Xanten in winter) to 48 °C (Carnuntum).

The tank is the most speculative part of the bath. Where an aqueduct reached a town, water came by pipe; small baths often managed with wells and cisterns (Yegül 2013, p. 11). The bath at Apollo Hylates had no piped water: “no direct supply of running water was found”, Scranton reports, and the water “must have been brought in jars” (Scranton 1967, p. 61). The tank and pipes in the model show one way a small bath could be supplied. At Kourion the other way was carrying.

A bath could not be lit in the morning and used at noon. Its floors and walls are tonnes of masonry, and they take days to charge. At Xanten, Grassmann needed six days in winter and seven to eleven in summer before the daily fuel settled to a steady rate; after that the inside held its temperature whatever the weather did (Goulianou 2020, pp. 42–43). The Sardis team allowed three days to warm the building and a whole day for the water.

The fuel figures spread widely. Hüser’s single room took 1.6 kg of wood an hour. The Xanten caldarium burned 120 kg a day at first in winter and 100 kg after six days, holding the room at 31 °C; in summer it came down to about 65 kg. The Sardis bath burned about 15 kg of oak an hour, some 360 kg a day; its builders reckoned that with drier wood and the changes they had learned to make, 6 kg an hour would have served (Yegül & Couch 2003, via Goulianou 2020, pp. 43–44).

Our model burns 4.2 kg of wood an hour, about 100 kg a day, from the moment of lighting, on a winter day of 12 °C outside. After the first day the caldarium air has reached about 25 °C, its floor 37 °C and the tub 27 °C. By the end of the third, with 300 kg of wood gone, the bath is close to steady, and on the fourth the caldarium air holds at about 35 °C. Measured caldarium air runs at 30–35 °C in the Catania study, 31–37 °C at Xanten and about 32 °C at Carnuntum; the Sardis team aimed for 40–43 °C, and a computer model of its hot room put the mean at 35 °C, with the hottest air trapped in the vault (Oetelaar 2013). Our numbers fall inside the measured ranges because the model was tuned to put them there. What they add is the shape of the warm-up, and they are an argument, not a measurement.

Once warm, a bath cost less to keep warm than to warm again, and the evidence points to fires kept in day and night, or banked overnight and stoked hard in the early morning (Goulianou 2020, p. 38). When our stoker stops at the end of the fourth day, a day later the caldarium air has fallen to about 23 °C, the floor to 28 °C and the tub to 27 °C. We know of no measured cooling curve from a rebuilt bath.

The tepidarium has no fire of its own. Its gas has already passed under the caldarium: in the model about 40 per cent of it carries on through 3 arched openings in the partition below the floor, and the rest goes up the caldarium’s walls. Our tepidarium air settles at about 26 °C. The only figures to compare come by analogy with Turkish baths, whose warm rooms run at 23–25 °C (Brödner 1983, via DeLaine 1988), which Maréchal warns can mislead.

Builders set the heat of a room by what they put in its walls. At Bearsden one warm room had no wall lining, only three or four wall flues (Breeze 2016, pp. 59–61). At Apollo Hylates the steps are spelled out: the tepidarium, with a hypocaust under its north half only, one wall flue and no hollow walls, was “only slightly warmed”; the sweat room beyond, with hollow walls on three sides, “distinctly warm”; the caldarium next to the fire, “genuinely hot” (Scranton 1967, pp. 58, 61).

Vitruvius wanted hot rooms vaulted in masonry. Under a timber roof, he says, a double ceiling is better, because the steam then wanders between the two shells and cannot rot the timber (V.10.3). In Cyprus, Harvey argues, builders preferred vaults of cut stone to the brick and concrete of Italy, though the direct evidence is thin: ashlar voussoirs at Apollo Hylates, a small niche vault of sandstone blocks at Agios Georgios, and the great ashlar vaults of the Salamis gymnasium baths lying where they fell into the hypocaust (Harvey 2024). The excavators of the House of Orpheus proposed vaults of wooden planks; Harvey finds that unlikely, given the fire and the damp.

Our hot rooms share one barrel vault of cut limestone, springing 2.7 m above the floor across the 3.8 m width, with the roof tiles bedded on its back. A flat terrace over the vault is possible, and the question is open. The cold rooms have a timber gable roof. The model needs about 1,350 roof tiles, and about 550 tonnes of stone and mortar for its walls.

Seneca recalled that “our ancestors thought nothing hot unless it was dark” (Seneca, Letters 86.4). Our hot rooms keep to that habit, with one small south window each, glazed in blue-green cast glass. The “quantities of broken glass panes” at Apollo Hylates show that small Cypriot baths were glazed (Scranton 1967, p. 58).

The commonest drawing of a hypocaust shows flames licking along under the floor. The fire stayed in the mouth of the furnace, and what ran under the floor was hot gas. Statius, describing the baths of Claudius Etruscus in about AD 90, found the words for it: a “languid fire” that wanders through the building, where “the hypocausts roll out a thin vapour” (Statius, Silvae 1.5.57–59).

The hot room was humid, and it was never a steam room. Its damp came from the tub and from water splashed on hot floors, and humidity in rebuilt hot rooms runs at 60–70 per cent (Goulianou 2020, pp. 42–43). Nobody piped steam into it. The dry sweat room was a separate room, the laconicum, round and domed in Vitruvius’s account, with a bronze shield hung on chains in an opening at the top of the dome, raised and lowered to set the heat (V.10.5). Our small bath has none.

Three smaller errors go with these. The box tiles are often taken for water pipes; they carried smoke, and the water ran in lead. The brick stacks at a ruin are often taken for the floor; they held it up. And Sergius Orata, “the first to build hanging baths” in about 95 BC according to Valerius Maximus (9.1.1), popularised a device that Fregellae and the Stabian Baths at Pompeii already had (Fagan 1996).

Hot bathing on the island is older than Rome. At the palace of Vouni a heating room was built in the second phase, about 475 BC, to serve a bath. Gjerstad, who dug it with the Swedish Cyprus Expedition, thought the heat rose inside hollow walls to a sweat room above and called it “the earliest known hot bath” (Gjerstad 1977, p. 102; Hellström 2009, p. 4). It is a heated chamber under a room, an oddity more than an ancestor. In the Hellenistic period Cypriots bathed in the Greek way, in separate hip-baths round a circular room: at Amathus a rotunda on the agora held at least twenty-eight, and the type is known at Kition too (Lemesos 1998, p. 52). Whether it was built early in the period or in the later 2nd century BC is disputed.

Baths with hypocausts came with the early empire. Harvey puts the Salamis gymnasium baths under Augustus, and the large baths on the Kourion acropolis possibly in the early 1st century AD (Harvey 2024). Salamis was rebuilt under Trajan and Hadrian after the earthquake of AD 76/77, with a caldarium of 29 by 13.70 m under a barrel vault of large stone blocks and a central sweat room fed by two furnaces 2.50 m below its floor (Karageorghis 1969, pp. 186–191); the architecture course walks through it. The Apollo Hylates bath went up in about AD 101/2, the Amathus agora baths in the 1st or 2nd century according to Harvey or the 4th according to the Limassol history (Lemesos 1998, p. 63). Rich houses at Nea Paphos and Kourion had their own.

The earthquakes of the 4th century ended many of them. Salamis came down in 332 and 342 and lay abandoned for about a century before it was rebuilt as the baths of Christian Constantia; the Kourion acropolis baths were abandoned in the 4th century (Herscher 1995), and the Apollo Hylates bath turned to other uses (Scranton 1967, p. 56). The Earthquake House at Kourion fell in the same century. Christian Cyprus went on bathing. Eustolios gave Kourion new baths in the early 5th century, with mosaic floors and a cross-shaped hot room facing south-south-west for the winter afternoon sun (Soren & James 1988, pp. 233–234), and Agios Georgios near Pegeia had a bath in the 5th or 6th. The Arab raids from 649 ended town life at Salamis, Amathus and Kourion, and we know of no Cypriot bath in use after the 7th century (see the Roman period).

A visitor sees the system from below. The floors fell into the hypocaust or were broken up for their tiles and marble, and the bronze and lead were taken long before. What remains is the hypocaust floor and its pilae, some short or missing, the lower courses of the walls, sometimes the mosaics of the cold rooms and an empty plunge. The model’s ruin state shows our bath that way.

In the Republic of Cyprus the place to see a hypocaust is the Baths of Eustolios at Kourion, under the modern roof, where the hypocausts of the warm and hot rooms, the firing chambers and the built-in hot basins can be seen on the west side. The acropolis baths by the nymphaeum can be seen on the same visit, and the Apollo Hylates bath beside the sanctuary west of the city. At Amathus the agora baths lie beside the re-erected Doric portico, and near Pegeia the bath at Agios Georgios stands beside its three basilicas. The House of Orpheus bath at Nea Paphos is not, so far as we know, on the usual visitor route.

The Salamis gymnasium baths, with their exposed hypocausts and the niche mosaics of Leda and the Niobids, are in the north of the island, under Turkish occupation since 1974. The Department of Antiquities has not excavated there since; Ankara University has dug there from 1998/99, work that is contested and illegal under international law. Vouni lies in the north too.

Much of the bath drawn here is firm: the order of the rooms, the hypocaust, the pilae, the floor and both kinds of hollow wall are excavated types, with Cypriot examples published. The furnace channel, the testudo and the boiler are inferred, the outlets reconstructed, the tank speculative.

Some gaps are in the record itself. We know of no analysis of fuel from a Cypriot furnace. Scranton’s text gives no height for the Apollo Hylates pilae, so the House of Orpheus figure is the only Cypriot one we have. We know of no measurement of the gas at a bath’s outlets, and of no published curve for how fast a rebuilt bath cools; the model’s figures for both are guesses with arithmetic behind them. Whether Cypriot hot rooms were tiled over their vaults or carried flat terraces is unresolved.

The heat model, though tuned to the experiments, runs hotter than the gas measured under real floors and evener than the floors measured above it. The measured rooms are less tidy. At the Saalburg in 1979, with the fire in for five days, the floor beside the furnace stood at 38 °C, and the far corner of the same room at 9 °C.