Metonic Cycle

The coincidence at the heart of it

Two clocks run the sky. One is the year — the time the Sun takes to return to the same point in the seasons, 365.24219 days. The other is the lunar month — the time from one new Moon to the next, 29.530589 days. These two periods have nothing to do with one another physically; the Earth's orbit around the Sun and the Moon's orbit around the Earth are governed by independent dynamics. So in general the Moon's phases drift relative to the calendar: a full Moon on your birthday this year will fall about eleven days earlier next year, because twelve lunar months (354.37 days) fall short of a year by almost eleven days.

Now do the arithmetic over a longer span. Divide the year by the month: 365.24219 / 29.530589 = 12.36827 lunar months per year. That fractional part — 0.36827 — is the problem. It refuses to be a simple fraction. But multiply by 19, and something remarkable happens: 19 × 12.36827 = 234.997. Almost exactly a whole number. Nineteen years contain very nearly 235 whole lunar months, falling just three-thousandths of a month short. That near-perfect closure is the Metonic cycle, and it means that after 19 years the Moon comes back to the same phase on the same date — the full Moon of this June will be a full Moon again 19 Junes from now.

The two periods, to the day

The whole cycle hinges on how close two independently measured durations come to each other. Here they are, computed from the modern mean values:

19 tropical years  = 19 × 365.24219 d  = 6939.6016 d
235 synodic months = 235 × 29.530589 d = 6939.6884 d
                                          ----------
difference         =                       0.0868 d ≈ 2.08 hours

Two hours of mismatch across nearly nineteen years. As a fractional error that is 0.0868 / 6939.6 = 1.25 × 10⁻⁵ — about one part in 80,000. Few naturally occurring period ratios in the Solar System are approximated by small integers this well. It is the astronomical equivalent of discovering that a guitar string and a piano string happen to be in near-perfect tune across nineteen octaves.

There is a second, deeper coincidence hiding inside the first. Nineteen tropical years also contain very nearly a whole number of sidereal months (the Moon's orbit relative to the fixed stars, 27.321662 days):

6939.60 d / 27.321662 d = 254.0 sidereal months   (exactly 235 + 19)

The relation 254 sidereal = 235 synodic + 19 is no accident: in 19 years the Moon laps the stars 254 times, and laps the Sun 254 − 19 = 235 times, because the Sun itself goes around the sky 19 times. So the same single coincidence ties together the year, the synodic month, and the sidereal month all at once.

Why calendars insert seven leap months

A lunisolar calendar wants two things at once: months that begin at new Moon (so the date tells you the lunar phase) and years that stay locked to the seasons (so spring stays in spring). Twelve lunar months is too short by about 11 days a year, so a pure 12-month lunar calendar — like the Islamic calendar — slides backward through the seasons, completing a full loop in about 33 years. To keep the seasons fixed you must occasionally insert a thirteenth month. The Metonic cycle tells you exactly how often:

235 months over 19 years = 19 ordinary years × 12 months + 7 extra months
                          = 228 + 7
so 12 years have 12 months, 7 years have 13 months.

Seven intercalary (leap) months per 19-year cycle. In the standard Babylonian and Hebrew schemes the leap years are years 3, 6, 8, 11, 14, 17, and 19 of the cycle. The Hebrew calendar still uses exactly this rule today, adding a second month of Adar (Adar I) in those seven years. The result is a calendar whose months track the Moon to within a day and whose Passover always lands in spring — a 2,400-year-old piece of working number theory.

Metonic, Saros, and the alternatives

The Metonic cycle is one of several integer resonances ancient astronomers tried. What sets it apart is which periods it closes. A cycle is only useful if the quantity you care about (phase, eclipse, season) returns. Here is how the main candidates compare:

Cycle	Length	Synodic months	What recurs	Residual error
Octaeteris	8 years	99	Phase (crudely)	~1.5 days / cycle
Metonic	19 years	235	Phase on calendar date	~0.086 day / cycle
Callippic	76 years	940	Phase, sharper	~0.26 day (Metonic ×4 − 1 d)
Hipparchic	304 years	3760	Phase, sharpest	~0.06 day / cycle
Saros	18 yr 11.3 d	223	Eclipses (same geometry)	closes draconic + anomalistic, NOT the year
Exeligmos	54 yr 33 d	669	Eclipses at same longitude	3 × Saros, integer days

The crucial contrast is Metonic versus Saros. Both involve the synodic month, but they aim at different targets. The Metonic cycle locks the synodic month to the tropical year, so phases land on the same dates. The Saros cycle locks the synodic month to the draconic month (the 27.212-day cycle of the Moon crossing the ecliptic, which governs eclipses) and the anomalistic month (27.555 days, controlling apparent size), so eclipses repeat with the same character. But the Saros is not a whole number of years — it is 18 years and about 11 days — so a Saros eclipse falls 11 days later in the calendar each time and shifts roughly 120° west in longitude. Metonic returns the calendar; Saros returns the shadow.

Stacking cycles for more accuracy

The Greeks knew the Metonic cycle was not perfect and built better ones by stacking it. Callippus of Cyzicus noticed around 330 BC that four Metonic cycles overshoot by almost exactly one day:

4 × 6939.69 d (235 months each) = 27758.75 d
76 tropical years               = 27758.40 d
overshoot                       = 0.35 d, so drop 1 day:
Callippic cycle = 76 years = 940 months = 27759 days (integer!)

The Callippic cycle of 76 years made the average year 365.25 days exactly — the value later baked into the Julian calendar. Hipparchus, two centuries on, found that four Callippic cycles (304 years) still ran a touch long and trimmed another day, producing a 304-year, 3,760-month cycle good to about an hour and a half per 19 years. Each refinement is the same Metonic idea pushed to a longer baseline, where the residual two-hour-per-cycle slip can be averaged out by adding or dropping whole days.

The Golden Number and the date of Easter

The longest-lived application of the Metonic cycle is the Computus, the algorithm the Christian church uses to find the date of Easter. Easter is defined as the first Sunday after the first ecclesiastical full Moon falling on or after the spring equinox (fixed at March 21). To find that full Moon without nightly observation, the church assumes lunar phases repeat exactly every 19 years — pure Metonic logic. A year's place in the cycle is its Golden Number:

Golden Number = (year mod 19) + 1        →  a value 1 … 19

Example, year 2026:  2026 mod 19 = 12  →  Golden Number 13
Example, year 2025:  2025 mod 19 = 11  →  Golden Number 12

The Golden Number indexes a table of epacts — the age of the Moon in days on January 1 — from which the ecclesiastical full Moon, and hence Easter, is read off. Because the underlying Metonic assumption drifts by about a day every 219 years, the medieval Julian Computus slowly lost sync with the real Moon. The Gregorian reform of 1582 added two correction terms — the "solar equation" (for dropped leap days) and the "lunar equation" (which advances the epact 8 times in 2,500 years) — to keep the calculated full Moon tracking the true one. The Metonic cycle is therefore not just historical trivia: a version of it, patched with corrections, still sets the date of Easter every year.

Carved into the Antikythera mechanism

The most spectacular physical embodiment of the Metonic cycle is the Antikythera mechanism, a bronze geared computer recovered from a Roman-era shipwreck and dated to roughly 100–150 BC. On its back face is a large spiral dial divided into 235 cells — one Metonic month per cell, five turns of the spiral covering the full 19-year cycle. A subsidiary dial inside it tracks the 76-year Callippic cycle, and another tracks the 223-month Saros for eclipse prediction, with a 54-year Exeligmos correction dial beside it.

The gearing that drives the Metonic pointer is a direct mechanical statement of the resonance: a train of bronze gears converts 19 turns of the year-wheel into 235 turns of the month indication, using tooth counts (such as a 53-tooth gear) chosen to realise the ratio 235/19 exactly. It is, as far as we know, the oldest surviving analog computer, and its central function is to compute the very coincidence this page is about — proof that the Metonic cycle was not abstract astronomy but engineered, hand-cranked technology more than two thousand years ago.

Seeing it for yourself

The cycle is observable on a human timescale, which is part of its charm. Pick any sharply dated lunar event and look 19 years on. The blue Moon, the harvest Moon, a notable lunar eclipse — all tend to recur near the same calendar date 19 years later. A worked instance: a full Moon on June 20, 2016 was followed by a full Moon on June 21, 2035 (one day later, exactly the expected two-hour-per-cycle slip rounded up across the interval). Birthday-Moon coincidences work too: if you were born under a first-quarter Moon, the Moon was again at first quarter, within a day, on your 19th and 38th birthdays.

The slip is what makes long-term observation interesting. Two hours per cycle is invisible in a single 19-year jump, but it compounds. Over the 304-year Hipparchic span it has grown past a day and the calendar date of a given phase has visibly walked. This is the same drift that forced the Gregorian lunar-equation patch, and it is why no purely Metonic calendar can run untended for millennia.

Common misconceptions and edge cases

• It is not exact. "19 years = 235 months" is an approximation good to about one part in 80,000, not an identity. The two-hour residual per cycle is real and accumulates to a full day in roughly 219 years.
• It governs phases, not eclipses. A common confusion is to expect eclipses to repeat every 19 years. They don't — that's the Saros (223 months, 18 years 11 days). The Metonic cycle does not close the draconic month, so the Moon is generally not at a node on the matching date and no eclipse occurs.
• Tropical, not sidereal, year. The cycle is built on the tropical year (equinox to equinox, 365.24219 d), because that is what keeps seasons fixed. Using the sidereal year (365.25636 d) gives a slightly different and less calendar-useful closure.
• The Hebrew calendar's year is slightly too long. Because it uses a fixed mean synodic month of 29 d 12 h 793/1080 h that is about 0.6 seconds too long, the Hebrew calendar drifts later relative to the seasons by roughly one day every 216 years — a direct descendant of the Metonic residual.
• The Golden Number is off by one if you forget the "+1". The historical convention numbers the cycle 1–19, so Golden Number = (year mod 19) + 1. Dropping the +1 shifts every epact lookup and gives the wrong full Moon.
• "Blue Moon" frequency is a Metonic by-product. Because 235 months span 19 years, there are 235 − 19×12 = 7 "extra" full Moons per 19 years — which is exactly why a second full Moon in a calendar month (a blue Moon) happens about 7 times in 19 years, or once every ~2.7 years on average.