Dresden Codex · pages 51–58 · candidate bases 1083 and 1116
The Maya eclipse table, run to your year
On eight pages of a Maya book, a table of 69 new moons marks the days a solar eclipse might come. It covers 405 months; then a new table has to begin. Choose where the next one begins, pick a year up to 2967, and see how many of its warnings still match real eclipses.
the table’s warningsplaceholdersNASA solar eclipseeclipse no station matches
Sixty-nine new moons
The Dresden Codex is one of four Maya books known to survive. It is a screenfold of bark paper, acquired in Vienna in 1739 for the library in Dresden. On pages 51 to 58 runs a table of 69 dates, each written as a running count of days from a base date in the 260-day calendar, 13 Muluc. Each date falls on a new moon. The count steps by 177 or 178 days, six lunar months, or by 148 days, five months, and stops at 11,959 days: 405 months.
In 2025 John Justeson and Justin Lowry set out how the table was built. Fifty-five of the dates come in runs of six (once seven), six months apart. They read these as warnings: new moons close enough to a lunar node for a solar eclipse to be possible. The other 14 are placeholders. They bridge the jumps of 11 or 17 months between one run and the next, because six lunar months (177.18 days) are a little longer than the average 173.31 days between eclipse seasons. The whole table, in the Maya numerals of its running counts:
Bars count five, dots count one, the shell is zero, and each position up is worth twenty times the one below, except the third, worth 360 days. The codex has several copying slips. In one, the first count is written 7.17 (157 days) where 8.17 (177) is meant; in another, the fourth reads 1.15.14 where 1.15.19 (679) is meant. The numerals above are the corrected counts. Click one to see its station in the chart.
Why 405 months
Four hundred and five lunar months come to 11,959.89 days, a fraction short of 11,960. And 11,960 days is exactly 46 rounds of the 260-day calendar. With the table, Maya calendar specialists could keep the moon and the ritual calendar in step for 33 years.
The eclipses are another matter. Sixty-nine eclipse seasons take about 11,958.3 days on the paper’s figures. If a new table starts on day 11,960, its base sits about a day and a half later relative to the node than the old base did, and each further restart adds about a day and a half more. Stations at the edge of each run slide out of the eclipse window; placeholders slide in. In the paper’s worked example of a restart on day 11,960, four warnings already sit past the window and three placeholders inside it.
In the chart above, choose “after 405 months” and run the centuries. The stations drift right, off the shaded band where NASA’s eclipses sit.
Where the next table starts
Justeson and Lowry propose that the next table began before the current one ended. Two earlier stations were better starting points: month 358, where 358 lunar months sit about a tenth of a day from 61 eclipse seasons, and month 223, the cycle now called the saros, which misses by about four tenths of a day the other way. Restarting four times at 358 and once at 223 spans 1655 months. In whole days, with the station at month 358 and the saros rounded to 6,585 days, that is 4 × 10,572 + 6,585 = 48,873 days. On their figures this keeps the table in step for a few millennia. It is a hypothesis; no successor table survives.
NASA’s catalogue of solar eclipses, compiled by Fred Espenak and Jean Meeus, lets each rule be tested. Each table’s 56 warning months (the base and the 55) are matched to the catalogue by lunation number. From either candidate base:
| Restart at | Mean shift per restart (paper) | Mean shift (173.3100 d) | Matches in 1500 | 2000 | 2967 |
|---|---|---|---|---|---|
| 405 (the end) | +1.563 d | +1.499 d | 25 | 0 | 0 |
| 358 | +0.098 d | +0.041 d | 56 | 56 | 56 |
| 223 | −0.423 d | −0.459 d | 39 / 45 | 5 / 11 | 0 |
| 4 × 358, 1 × 223 | −0.006 d | −0.059 d | 56 | 56 | 52 / 54 |
Matches out of 56, from the 1083 base / the 1116 base (one figure when they agree). The shift is the mean per restart, in days; plus means the next base falls later relative to the node.
Restarting after 405 months drops below 50 matches in 1312 and reaches zero in 1770. Month 223 alone drops below 50 about as early (1299 or 1368, depending on the base) and reaches zero in 2183 or 2306. Month 358 alone keeps 55 or 56 matches in every table to 2967, and the four-to-one mix keeps at least 52. On this list, 358 alone drifts least. The paper prefers the mix because, with its eclipse-season length of 173.30906 days, the mix reduces the average drift. With the 173.3100 days used for the chart, the half eclipse year that fits NASA’s eclipses, 358 alone drifts less.
Which years
The 69 counts give intervals from a base; where that base falls in history is debated. To place it, Justeson and Lowry look for solar eclipses they judge observable in Maya territory on the table’s base day (13 Muluc, position 169 of the 260-day count) or its last day (12 Lamat, 168), or a day earlier. They find four, in two pairs exactly 405 months apart. NASA’s catalogue confirms all four and the spacing within each pair (dates in the Julian calendar):
| NASA eclipse (Julian date) | Type | Lunation | Days within the pair | 260-day position (584,285) | (584,283) |
|---|---|---|---|---|---|
| 9 June 1043 | total | -11,831 | — | 168 | 170 |
| 7 March 1076 | total | -11,426 | 11,960 | 168 | 170 |
| 19 April 1083 | annular | -11,338 | — | 167 | 169 |
| 16 January 1116 | annular | -10,933 | 11,960 | 167 | 169 |
The 260-day positions depend on how the Maya calendar is tied to ours, through a single constant, the correlation. With 584,285 the four eclipses land on positions 168, 168, 167 and 167, the positions the paper gives. With 584,283, another widely used value, the 1083 and 1116 eclipses fall on 169, the base day itself, and the other two on 170: the paper’s positions are not reproduced. The paper favours 1083 and 1116 as the base; switch between them above.
What stays open
NASA’s catalogue lists solar eclipses visible from somewhere on Earth. Most new moons near a node bring one somewhere, so a warning month that contains an eclipse does not mean the eclipse was visible in Yucatán. The chart tests the table’s rhythm against the nodes; visibility in the Maya region is a separate question.
The base year is not settled: the paper favours 1083 or 1116 and doubts 1043. Restarting at 358 or 223 is a hypothesis drawn from the table’s structure and from eclipse records. The placeholders are read as placeholders from the pattern of the intervals. In the paper’s geometry, with the base 0.11 days before the node, one warning, day 10,040, sits 12.0 days from the node, just past the window.
The chart uses mean new moons; its dates are mean new-moon dates. A real new moon can fall up to about 14 hours either side of the mean one, which can move it to the next or previous day.
Sources
The station counts are taken from the corrected table in Justeson & Lowry (2025), Fig. 1. Every figure on this page is traced in the project’s source notes.
- John Justeson and Justin Lowry, “The design and reconstructible history of the Mayan eclipse table of the Dresden Codex”, Science Advances 11 (43), eadt9039, 2025, doi:10.1126/sciadv.adt9039; full text, CC BY.
- Fred Espenak and Jean Meeus, Five Millennium Canon of Solar Eclipses, NASA, 2006; solar eclipse catalogue, fourth to thirtieth centuries (eclipse predictions by Fred Espenak, NASA GSFC).
- Floyd G. Lounsbury, “Maya numeration, computation, and calendrical astronomy”, Dictionary of Scientific Biography 15, 1978, pp. 759–818 (the ratio of 1447 days to 49 months).
- SLUB Dresden, The Dresden Maya Codex: material, the 1739 acquisition, the eclipse pages; Dresden Codex (Wikipedia).
- Jean Meeus, Astronomical Algorithms, 2nd ed., 1998, ch. 49: the mean new moon (k = 0 on 6 January 2000), used for the new-moon dates in the chart.