The Voynich Manuscript: A Book Nobody Can Read

A small book, about the size of a modern hardback (roughly 23 by 16 centimetres), lies in a climate-controlled vault in New Haven. It is bound in limp vellum. Inside are roughly 240 surviving pages of calfskin covered in a flowing, confident script that nobody alive, and as far as anyone can prove nobody dead, has ever been able to read. Beside the writing are plants that match no plant exactly, circles of stars and naked women, a zodiac with Latin-alphabet month names added by a later hand, green pools linked by tubes, and a folding sheet of nine rosettes joined by causeways. The book is catalogued as Beinecke MS 408, "Cipher manuscript". Everyone else calls it the Voynich Manuscript, after the antiquarian bookseller who brought it to light in 1912.

This entry sets out what can be documented about the object (its material, its date, its owners, its structure) and keeps that apart from what has been claimed about it. Many people claim to have solved it, and each claim is examined below. The page scans on this entry come live from Yale's own image server. You can magnify any folio, sharpen faded ink, cut out a region at full resolution, and load the standard scholarly transliteration to test a decipherment of your own.

The object: what the manuscript physically is

The manuscript is a codex of vellum (calfskin) leaves gathered into quires. The quires carry their own numbering, and the gaps in that numbering show that some leaves are missing. Several leaves are foldouts, the largest of them the multi-panel sheet usually called the "Rosettes" page. The text is written in iron-gall-type brown ink. The drawings are outlined in the same ink and coloured with paints that are often thought to have been added later and less carefully than the outlines. The foliation (the folio numbers "f1r", "f68r3" and so on, used by everyone who works on it) was added afterwards in Arabic numerals. That is why modern references use foliation rather than page numbers.

In 2009 the University of Arizona radiocarbon laboratory, working with Yale, dated four samples of the vellum from different leaves. The results were mutually consistent: the animals whose skins made the parchment died between 1404 and 1438 at 95% confidence. That dates the parchment, not necessarily the writing. In principle old blank vellum could be written on later, and some sceptics have argued exactly that. Ink and paint analyses made around the same time (reported by McCrone Associates) found nothing inconsistent with that period. Most specialists now treat an early-fifteenth-century origin for the book as the working assumption, and the date rules out older theories outright: Roger Bacon (d. 1292) cannot be the author, even though that attribution drove interest in the book for most of the twentieth century.

Its owners: a documented chain with gaps

The chain of ownership is unusually well documented for a medieval book, but it is not complete:

• Before c. 1600, nothing is known. Where and by whom the book was made is entirely inferred.
• Jacobus Horčický de Tepenec (d. 1622), a chemist and physician in Prague, signed the first folio. The signature was later erased but can be recovered under ultraviolet light, and it reads as his ennobled name. He received his title in 1608, so the signature is no earlier than that.
• Rudolf II, Holy Roman Emperor (r. 1576–1612), is said to have bought the book for 600 ducats. The claim rests on a letter by Johannes Marcus Marci of Prague, who reports it as hearsay from another man. It is plausible and often repeated, but it is second-hand, not a receipt.
• Georg Baresch, a Prague alchemist, owned it by the 1630s and wrote to the Jesuit polymath Athanasius Kircher in Rome asking for help reading it. No reading of the book by Kircher is known.
• Marci inherited it and sent it to Kircher with a covering letter dated 1665 or 1666. The letter is still with the manuscript.
• Through the Jesuit libraries it ended up among books sold from the Villa Mondragone at Frascati, where Wilfrid Voynich acquired it around 1912.
• After the deaths of Voynich (1930) and his widow Ethel (1960), the dealer H. P. Kraus bought it in 1961, failed to sell it, and gave it to Yale in 1969.

The structure: six "sections" that are modern labels

The section names used everywhere are modern descriptive labels based on the pictures. The book itself names none of them:

• Herbal: single plants with roots, leaves and flowers, each with a block of text. A few have been tentatively matched to real species, but most are composites or unidentifiable.
• Astronomical / astrological: circular diagrams of suns, moons and stars, including a zodiac sequence with small female figures, many holding stars, arranged in rings. The month names written beside them in Latin script (e.g. abril, may) are in a later hand. Whether their spelling points to Occitan, Catalan, French or another Romance form is disputed.
• "Balneological" / biological: naked women in pools and tubs joined by branching tubes that look to some like plumbing and to others like anatomy.
• Cosmological: large circular diagrams, including the Rosettes foldout.
• Pharmaceutical: drawings of plant parts next to jar- or container-like objects.
• Recipes / "stars": pages of short paragraphs, each marked with a small star, with no other illustration.

The script: what "Voynichese" looks like

The script has roughly two dozen common glyphs and a scatter of rare ones. For computer work, researchers use a standard Latin-letter transliteration alphabet, EVA (the Extensible or European Voynich Alphabet), devised in the late 1990s by Gabriel Landini and René Zandbergen. In EVA a typical line reads like daiin shedy qokeedy okar chol. EVA does not claim what the glyphs mean. It only records their shapes so that everyone counts the same thing.

Several features stand out:

• Word structure is rigid. Words divide into something like a prefix, a core and an ending, with strong rules about what can follow what. The EVA glyph q appears almost only at the start of a word, usually followed by o.
• Repetition. The same word, or near-variants of it, often appears two or three times in a row (sequences like qokeedy qokeedy qokedy). That is unusual for natural-language prose.
• Position effects. Certain tall glyphs (the "gallows", EVA k, t, p, f) cluster at the start of paragraphs and lines, and some glyphs prefer line-ends. The line seems to work as a unit of its own.
• Low character-level entropy. Measured glyph by glyph, the text is more predictable than typical European alphabetic languages. William Ralph Bennett reported this in the 1970s, and later work has repeated it, though the exact figures depend on how glyphs are split in the transliteration.
• Two "languages". Prescott Currier, an American cryptanalyst, showed in 1976 that the text falls into at least two statistical dialects, Currier A and B, which match differences in handwriting and in page groups.
• Several scribes. In 2020 the palaeographer Lisa Fagin Davis argued from letterforms that five scribes wrote the manuscript. That makes it the product of a workshop or collaboration, not one eccentric person.
• Word statistics that behave like language. Word frequencies roughly follow Zipf's law. Montemurro and Zanette (2013) found that some words cluster by section in the way content words do in real texts.

In 2024, multispectral images of ten pages, taken by the Lazarus Project in 2014 and released with permission, let Lisa Fagin Davis recover erased Roman-letter writing in the margin of folio 1r. It is most plausibly an early reader's attempt to set up a key or alphabet. It is evidence that people were trying to decipher the book long ago, not evidence of what the book says.

Symbology and encoding: what could be hidden here, and how it might be hidden

Taken as a system of signs, the manuscript raises every question this archive asks of sacred scripts. For each one we separate what is observed from what is inferred.

The script as object. The glyphs are written fluently, without the hesitations of someone copying unfamiliar shapes. That is observed. It suggests trained scribes working in a script they used regularly. Inferred, not proven: that the script encodes speech at all.

Letter-value and number systems. Some researchers note that certain glyphs look like fifteenth-century numeral shapes and Latin scribal abbreviations (the -9 "us/con" sign, the -ij ending). The resemblance is real and documented. Whether the glyphs function as those abbreviations is the core of several failed decipherments (see below).

Imagery as code. The zodiac circles hold 30 figures per sign in most places, which fits the degrees of a zodiac sign. That is a real, countable pattern and a strong hint that the astronomical pages concern days or degrees. The "nymphs" and pools have been read as balneology (bathing cures), as anatomy, as alchemy and as cosmology. All of these readings are interpretations. None is established.

The line as a unit. Line-initial and line-final effects are a documented, real structural feature. Some take them as a sign of a cipher, some of a notation or verse, some of a generated text. The feature is documented. What it means is open.

The decipherment claims, examined

A genuine decipherment must do three things. It must produce connected, grammatical text across paragraphs and pages, not a scattered list of words. It must use one consistent, fixed key that is not adjusted page by page. And it must predict things the decipherer did not already know, for example by reading a new page correctly, which other specialists can check. Every proposal so far fails at least one of these tests:

• William Romaine Newbold (1921), a Philadelphia professor, claimed the visible glyphs hid microscopic shorthand showing that Roger Bacon had a microscope. John Manly (1931) showed the "shorthand" was cracks in the ink. The carbon date later ruled out Bacon anyway. Rejected.
• William and Elizebeth Friedman and wartime study groups worked on it for years without a solution. William Friedman came to suspect an artificial "philosophical" language. That was a hypothesis, not a reading.
• Leo Levitov (1987) read it as the liturgy of a Cathar cult in a Flemish-based mixed language. Specialists found the language and the history untenable. Rejected.
• Gordon Rugg (2004) showed that a Cardan grille used with a table of syllables can produce gibberish with many Voynich-like properties. That argues a hoax is possible. It does not show that the book is one. Open. Later work, including Zandbergen's 2021 review of the grille approach, finds some properties hard to reproduce this way.
• Stephen Bax (2014) proposed sound values for some glyphs from supposed plant and star names, reading about ten words. It was offered as provisional, and he did not claim a full decipherment. Unconfirmed.
• Arthur Tucker and Rexford Talbert (2014) matched plants to New World species and proposed a Nahuatl-linked origin. That clashes with the 1404–1438 parchment date, which falls before 1492, unless the vellum was old when written on. Contested; generally not accepted.
• Greg Kondrak and Bradley Hauer (2016/2018) used computer methods that suggested the text might be Hebrew with anagrammed words. Hebrew specialists found the proposed sentences unconvincing. Unconfirmed.
• Nicholas Gibbs (2017) claimed in the Times Literary Supplement that it was an abbreviated Latin women's-health manual. Medievalists quickly showed that the proposed Latin did not make grammatical sense. Rejected.
• Gerard Cheshire (2019) claimed a "proto-Romance" language in the journal Romance Studies. Specialists rejected it as ad hoc, and the author's own university distanced itself. Rejected.
• Torsten Timm and Andreas Schinner (2020) proposed a "self-citation" method: a scribe makes each new word by copying and slightly changing a nearby earlier word. This reproduces key statistics of the text and supports the view that it could be meaningless but carefully made. Open. It is a serious model, but it explains how the text could be made, not that it was made this way.
• Claire Bowern and Luke Lindemann (2021), in the Annual Review of Linguistics, reviewed the statistics and concluded that the text has many language-like properties. They leaned towards the view that it records a natural language in an unusual writing system rather than a complex cipher. That is a judgement of likelihood, not a reading.

Our attempt, and how you can make your own. This archive has not deciphered the Voynich Manuscript, and it would be dishonest to present anything here as a translation. What we have built is a workbench on this page. It lets you do the same kind of testing the literature describes: load the Zandbergen–Landini transliteration, see the EVA text of whatever folio you are viewing, compute word frequencies, glyph entropies and line-position effects for that page or for the whole Currier A or B corpus, and apply a substitution key of your own to see what the text becomes. A proposal that turns one page into plausible words but the next into nonsense fails the fixed-key test. The workbench makes that visible in seconds. Any AI or human "decipherment" can be checked here against the three tests above.

Connections

The manuscript belongs to the world described in the archive's chapters on late-medieval scholasticism and natural philosophy (Scholasticism, ch28) and on the esoteric revival that later adopted it (Theosophy & the Occult Revival, ch37). Its Prague owners moved in the same circle of Rudolf II's court alchemy that preserved Hermetic texts such as the Emerald Tablet (Vault V03). It also belongs to a small family of unread or disputed scripts, including the Rohonc Codex (V11) and the Book of Soyga (V19).

From The Divine Archives · https://getconexto.com/vault/voynich-manuscript.html