Every Puzzle ProjectBlog

The first crossword series, rebuilt from the papers that copied it

The New York World printed a crossword every Sunday from December 1913, and its own pages survive badly. The papers that reprinted its FUN section kept a second copy of each puzzle. How we are putting the series back together from them, and what is still missing.

Alex DeJarnatt · Every Puzzle Project · 6 November 2026

The first crossword, Arthur Wynne's "Word-Cross Puzzle", ran in the New York World's Sunday supplement, FUN, on 21 December 1913. The puzzles that came after it, a new one every Sunday, are another matter. We know of no scans of the FUN section online; the few early pages researchers have reproduced come from sources they rarely name, most likely private collections and library microfilm. For most Sundays of the crossword's first four years, nobody today can say what puzzle the World printed.

But the World did not keep its puzzle to itself. Almost from the start, other papers carried the FUN section, crossword and all. The Minneapolis Journal printed the World's puzzle on 1 February 1914, six weeks after the first one. The Pittsburgh Press followed in June 1916 and the Tacoma Daily Ledger that August; by 1917 the Los Angeles Sunday Tribune, the Shreveport Times, the Fort Wayne Journal-Gazette, the New Orleans States, the Salt Lake Telegram and the Boston Globe were running it too. Each of those pages is a second copy of a World puzzle, and most of them are on Newspapers.com.

So we are rebuilding the series from its reprints. The record is the newyorkworld repository: one row per World Sunday, No. 1 to No. 211 (21 December 1913 to 30 December 1917), with every copy of each puzzle found in a partner paper, and a status page built from it.

The number in the grid

The World printed its series number inside the puzzle, in the shaded cross at the centre of its diamond grid. The Journal's first copy says No. 7; the next Sunday's says No. 8. One puzzle a week from No. 1 on 21 December 1913 fits exactly, which gives the record its one dating rule:

World No. n is expected on 21 December 1913 plus n−1 weeks.

Every World date in the record is an inference from that count, and the status page marks each one as such. What a page actually shows is recorded separately: the number printed in the grid, if any, and the paper's own date, which is a printing date, days to weeks after the World's. The Journal itself got the count wrong at first: its heading on 1 February 1914 said "No. 5" while the grid said No. 7, and the next week the heading was gone. The grid number is the one that counts.

The Journal ran each puzzle on the World's own date through No. 21 on 10 May 1914. Then it fell behind, missing Sundays and running two to three weeks late, and skipped a number altogether: No. 33 on 23 August, No. 35 on 30 August, and No. 34 never. The last number printed anywhere is No. 54, in the Journal of 17 January 1915. After that no copy in any paper carries one, so from 1915 the count is all there is.

And the count may be wrong. The Journal's Thanksgiving puzzle is No. 46, printed on 22 November 1914; the count puts No. 46 on 1 November. Its Christmas puzzle, No. 50, falls on 29 November by the count. A paper would not run a Christmas puzzle a month early, so either the Journal was weeks behind by then, or the World itself skipped some Sundays in 1914 and the count runs early from there on. One dated World page from the second half of 1914 would settle it. Until then the printed numbers stay beside the inferred dates.

Printings are not puzzles

The tables keep the two apart. A Boston Globe Sunday in 1917 is a printing of a World puzzle, not a Globe puzzle. A printing with a number in the grid is tied to that World number. A printing without one is tied to another copy by its first clue and the shape of its grid, and the record says which copy and how. From the two dates, the paper's and the inferred World date, each printing's lag is computed, so the page shows how far behind each paper ran: the Globe on the World's own date in 1917, Pittsburgh one or two weeks later, Tacoma two, Los Angeles anywhere from three to eight.

Indexing a page means reading only what identifies it: the heading, the number, the credit line, the first clue, and which solution the page carries. The puzzle text comes later, from OCR, into an xd file with its proof beside it.

Chains: the order, proved page by page

Each Sunday's page also prints the solution to the previous week's puzzle. That is the second kind of evidence, and it does not depend on the count. A link says that the solution on page B answers the puzzle on page A: the same grid shape, and answers that fit A's numbering, checked by eye on the scans. The two pages need not be in the same paper. Tacoma's solution of 28 January 1917 proves the puzzle Los Angeles printed six weeks later.

A chain is a run of puzzles in which every adjacent pair is linked. The order of the puzzles in a chain rests on what was printed; only its place on the calendar rests on the count. Two chains are held in place by the printed grid numbers: Nos. 7 to 33, and Nos. 35 to 52. The long one is 107 puzzles from 21 November 1915 to 30 December 1917, 106 verified links, and it is dated by a solutions box the World Magazine printed with the World's own dates on it, which Tacoma reprinted on 22 July 1917. The Journal's pages from January to July 1915 form shorter chains with no number on any of them, so their order is proved but their World dates are not.

A break in a chain is a puzzle that no held page answers. Some breaks are the World's own missing weeks: the Journal never ran No. 34, and No. 53 is in no paper either. Some are a partner's: a page that is not on Newspapers.com is recorded as missing from the scans, never as a skip by the World. And some close with one page: the last break before the solutions box, in February 1917, closed when a clipping of the Pittsburgh Press of 4 March 1917 turned out to carry the same puzzle as Shreveport's that day, and 107 puzzles became one dated run.

The chains have already corrected the count: the Journal printed one puzzle twice in the autumn of 1916, ran a week late for a month, then dropped a puzzle to catch up, and nine World dates moved. They also show where the method stops. From 1918 the Globe's solutions go on week after week, but its order is no longer the World's: a puzzle another researcher dates to October 1919 ran in the Globe in May 1920. So after 1917 the Globe's pages are good copies with no World date.

The Boston Globe's first crossword

The Globe printed its first crossword on 25 February 1917, on page 4, with no credit. The Tacoma Daily Ledger printed the same puzzle on 11 March 1917 as "Fun's Cross-Word Puzzle": the same grid, the same hand-lettered numbers 1 to 122, the same clues. So the Globe's first crossword was the World's, No. 167 by the count, and the Globe was one more FUN partner, running on the World's own date through 1917. The two copies differ only in typesetting: "clews" and "Mahometan" in Boston, "clues" and "Mohammedan" in Tacoma. The grid seems to have gone out as a ready-made plate or mat, and the clues as text that each paper set itself. For fourteen of the World's 1917 puzzles the Globe is the only copy found so far.

The record as of November 2026

211World Sundays, No. 1 to No. 211, Dec 1913 to Dec 1917
155have at least one copy
328copies, in 11 papers
64puzzles with copies in two or more papers
325paper-Sundays searched through 1917
PaperCopies fromPuzzles
The Minneapolis Journal1 February 1914103
The Pittsburgh Press11 June 191647
The Tacoma Daily Ledger27 August 191645
The Boston Globe25 February 191741
Los Angeles Sunday Tribune191632
The Shreveport Times191630
New Orleans States191721
Fort Wayne Journal-Gazette19164
Salt Lake Telegram19172
Seattle Post-Intelligencer19161
The New York World itselfNo. 4, from a reproduction1

Puzzles among Nos. 1–211 with a copy in each paper, counted from the repository's status table on 6 November 2026. A puzzle found in three papers counts once for each.

Of the 325 Sundays looked for in a given paper through 1917, 184 turned up a page, 105 found nothing, 34 fell on a Sunday the paper carried no FUN puzzle, and 2 are missing from the scans. "Nothing found" is not proof of absence: a text search misses pages whose type the OCR could not read, and not every such Sunday has been browsed issue by issue.

The 56 World Sundays with no copy fall into a few groups:

What would help most

The World's own FUN section for December 1913 and January 1914, Nos. 1 to 6. Four pages would close the gap at the start of the series, and they are most likely in a private collection, on the New York Public Library's microfilm of the World, or in another researcher's files. After that: a dated World page from the second half of 1914, to settle whether the count runs true; any paper that carried FUN between August and November 1915, to date the Journal's floating chains; and a copy of any Sunday still open on the status page, which closes it in every paper at once.

If you have access to one of these, please open an issue with the date and where you found it, with a clipping link if there is one. Please don't upload page images; issues are public. The status page shows every Sunday, what is held for it and from where, the chains, and the next batch to look for.

Sources