The Villa of the Papyri at Herculaneum is the only Roman-period library that has been physically recovered with its book collection intact. The building was a large seaside residence on the western edge of the Roman town of Herculaneum, looking out across the Bay of Naples toward Mount Vesuvius. It was buried in the 24 October 79 AD eruption of Vesuvius under a sequence of pyroclastic flows that reached approximately 400°C at the villa’s depth and that ultimately covered the structure to a depth of about 25 metres of consolidated pyroclastic material.
The heat carbonised the villa’s contents. The pressure entombed them. The combination preserved organic material (wooden roof beams, furniture, scrolls, food residues, even a few human bodies) that would otherwise have rotted away in any of the next nineteen centuries.
The library contains about 1,800 carbonised papyrus scrolls. Until 2023 they could not be read.
The villa
The villa was identified in 1750 by the Swiss-Italian military engineer Karl Weber, working on behalf of the Bourbon royal court of Naples, during the second wave of 18th-century Bourbon excavation at Herculaneum. The town had been rediscovered in 1709 when well-diggers in the modern village of Resina (above the buried Roman site) struck cut marble; the Bourbon excavation programme had been running intermittently since 1738 and was at the time the largest archaeological excavation anywhere in Europe.
Weber’s technique was tunnelling. The 25-metre depth of pyroclastic overburden made open-pit excavation impossible with 18th-century technology, so the Bourbon excavators drove narrow tunnels through the consolidated volcanic deposit and recovered Roman material in fragments as they encountered it. The Weber tunnels at the Villa of the Papyri ran for about 14 years (1750–1764) and mapped approximately 90% of the villa’s ground floor.
The villa was substantially larger than typical Herculaneum residences — about 250 metres of seafront frontage, several connected garden peristyles, an extensive sculpture programme that yielded approximately 90 marble and bronze pieces during the Weber tunnelling, and a private library room on the western wing. The owner has never been positively identified but the leading candidate (based on the sculpture programme and the library contents) is Lucius Calpurnius Piso Caesoninus — Julius Caesar’s father-in-law and the patron of the Greek Epicurean philosopher Philodemus of Gadara.
The scrolls came up in 1752–1754. They were found stacked in wooden bookcases along the walls of a single room about 4 metres square. The bookcases had collapsed and the scrolls had fused into a compressed mass of carbonised papyrus. The first Bourbon labourers to encounter them mistook the carbonised cylindrical objects for lumps of charcoal and threw several dozen of them into the spoil heaps before a passing supervising scholar recognised what they were.
The early reading attempts
The 18th- and 19th-century attempts to read the scrolls were heroic and largely unsuccessful. The carbonised papyrus had become brittle, the ink (carbon-based, not iron-gall) was the same colour as the carbonised papyrus matrix, and any attempt to physically unroll a scroll typically broke it into fragments. The Bourbon court established a dedicated workshop at the royal palace — the Officina dei Papiri — staffed by trained papyrus conservators who developed progressively more refined techniques for slow unrolling and transcription.
The standard 18th-century technique was the macchina di Piaggio, a mechanical device invented by the Genoese priest-conservator Antonio Piaggio that suspended the scroll horizontally and slowly unwound it onto a roller while a conservator transferred the unrolled portion to a backing sheet. The machine could unroll a scroll at about 1 centimetre per hour. Most scrolls broke during the process. Of approximately 800 scrolls subjected to the Piaggio method between 1754 and 1900, approximately 250 produced some readable text; the rest were destroyed or fragmented beyond recovery.
The readable scrolls turned out to be predominantly Greek philosophical texts of the Epicurean school — most importantly, previously-unknown works of Philodemus of Gadara (c. 110–35 BC), the philosopher who had probably been the library’s curator under the patronage of Piso. The Philodemus material included treatises on rhetoric, music, poetics, ethics, and theology that had been entirely lost to the medieval and early modern philosophical tradition. The Officina dei Papiri’s 19th-century editions of the Philodemus material are the basis for the modern study of Epicurean philosophy of the late Roman Republic.
A smaller fraction of the readable scrolls contained Latin material — including portions of an otherwise-lost poem by Lucius Calpurnius Piso that suggests the villa’s likely owner. No major canonical Latin or Greek work (no Cicero, no Virgil, no Aeschylus, no Sophocles, no Aristotle) has yet been identified in the readable portion of the collection. The library was apparently a specialised Epicurean philosophical collection rather than a general-purpose Roman aristocratic library.
About 1,200 scrolls remain unread. They are stored at the Naples National Library and at the Officina dei Papiri at the same institution.
The X-ray breakthrough
The decisive modern advance was the X-ray micro-CT scanning of intact scrolls by a University of Kentucky team led by the computer scientist Brent Seales through the 2000s and 2010s. Seales’s technique was to scan a sealed scroll at very high X-ray resolution (about 4 micrometres per voxel — fine enough to resolve individual papyrus fibres), then use a custom-built software pipeline to identify and segment the spiral wrap of the original scroll as a virtual three-dimensional surface, then unfold that virtual surface into a flat two-dimensional image.
The technique had a critical limitation: the carbon-based ink and the carbon-based carbonised papyrus had nearly identical X-ray attenuation, so the ink was invisible in the raw scan data. The first Seales successful application of the technique (to the En-Gedi Scroll in 2016 — a 5th-century AD Hebrew Leviticus scroll found in a Dead Sea cave) had worked because that ink was iron-based and had high X-ray contrast against the parchment matrix. The Herculaneum scrolls had no comparable contrast.
The Herculaneum breakthrough came from machine learning. The Seales team and a network of collaborating computer scientists developed a deep-learning model that could be trained to recognise the subtle micro-textural differences in the X-ray scan data between inked and uninked papyrus fibres — differences too small for a human reader to identify but consistent enough across millions of training pixels that a trained convolutional neural network could detect them.
The technique was made the basis of the Vesuvius Challenge — a public crowdsourced AI competition launched in March 2023 with $1 million in prize money funded by a consortium of Silicon Valley technologists. The competition’s first major success came in October 2023, when a 21-year-old University of Nebraska computer-science student named Luke Farritor trained a model that correctly identified the Greek word ΠΟΡΦΥΡΑϹ (porphyras, “purple”) in a previously-unread Herculaneum scroll fragment.
The full first scroll was read by the winning Vesuvius Challenge team in February 2024 — about 2,000 readable Greek characters from a previously-unknown Philodemus text on the Epicurean philosophy of pleasure. The Vesuvius Challenge has continued through 2024 and 2025 with progressively larger reading targets and progressively more sophisticated machine-learning pipelines. The pace of recovery is now approximately one new scroll every 4–6 months.
The expectation is that the readable corpus from the Villa of the Papyri will approximately double over the next decade as the X-ray technique is applied to the remaining 1,200 unread scrolls. The exact composition of the eventual recovered library is still unknown. There is a non-trivial possibility that the previously-unread portion contains canonical Greek or Latin works that have been entirely lost to the medieval transmission tradition — a recovered Aristotelian treatise, a recovered Aeschylean tragedy, a recovered Ennius poem. The probability is low but not zero.
The library has been waiting for nineteen centuries. The reading will take another decade or two.