Dating a piano

A piano's serial number was stamped by whoever actually built it — and in the piano trade that is frequently not the name painted on the front. Wikipedia lists 1,182 piano brand names against 222 manufacturers. Most of the difference is stencil pianos: instruments built by one firm and sold under another firm's name.

So the first question is not “what year is this number?” It is “whose number is it?” Look the serial up against the brand on the fallboard rather than the company that made the instrument, and you can get a precise, confident, wrong year.

Two makers in 222 publish figures complete enough to compute against. The rest of this page is about why that is, and what the others do instead.

Two tables that look identical and mean opposite things

Kawai's table and Grotrian's table have the same shape: a year, a serial number, ascending down the page. They are read in opposite directions, and nothing on either page warns you.

Bring Kawai's assumption to Grotrian's table and every answer comes out a year late. The difference is one word in a column heading, and it is the sort of thing a decoder gets wrong silently and forever. The convention is stored with each maker's data here rather than written once into the code.

What the makers publish

Kawai — four tables, and a warning attached

Kawai Musical Instruments Manufacturing Co., Ltd publish their serial tables on their own support site: Japan production from 1927, US production from 1988, Indonesian production from 2003, and a separate series for KX models.

Kawai attach three caveats. Verbatim:

“The serial numbers listed below are the approximate first number produced for the year shown.”

“Starting letters other than A or F should be disregarded.”

“Serial numbers for different models are not always sequential, so actual dates could vary.”

Read the third one twice. Kawai are saying that their own table does not settle the year. It gives the block a number falls in; the instrument's real date could differ. A tool that prints a bare year from this data is overstating what Kawai themselves claim, so this one does not.

Their US site refused us, and that was honoured. kawaius.com/robots.txt disallows our crawler by name and sets Content-Signal: ai-train=no. Nothing was taken from that host. Kawai's global site carries the same tables and carries no such restriction, and that is the one used here.

And the tables overlap. Japanese serials and KX serials are both plain digits, occupying the same numeric range at different times. A bare 20000 is 1954 if it is a Japanese-production number and 2001 if it is a KX — forty-seven years apart, from the same digits. The decoder showed every table a number could belong to rather than picking one, until Kawai's tables came off the page on 14 September. This one serial stays because it is the evidence: an overlap nobody can check is an assertion, and two years forty-seven years apart is not an answer anybody can use.

Grotrian-Steinweg — the table is theirs, and it is not here

GROTRIAN publish a table on their own service pages, in six languages, running from 1864 to 2010. At 144 rows it is the longest run of figures any maker on this page publishes.

This page used to carry it and no longer does. GROTRIAN’s terms name databases explicitly, and a table of 144 serial-to-year rows is a database by any reading. The site asks before it publishes rather than publishing and waiting to be told, and the letter is outstanding. Until it is answered the rows stay out, and the tool above says so rather than producing a year.

What is still worth knowing is the shape of it, because that is a fact about their numbering rather than a copy of it. Their rows give the last serial of each year, not the first — the opposite of Kawai’s, which is the trap this page was built around.

And three years are missing from it. Their table steps straight from 1945 to 1949: no row for 1946, 1947 or 1948. The company built in Braunschweig, and their published record of the immediate post-war period is simply not there. A decoder that walks the table without checking hands those numbers to 1949 without comment. This page used to report the range instead — and will again, if the answer is yes.

Steinway & Sons — keeps the records, runs its own lookup

Steinway do not publish a table. They operate a serial lookup of their own, and the database behind it sits on an authentication portal — a login, not a public page. Their own material says they can identify instruments from 1853 onward. Their public serial lookup page is a form that takes your details and directs you to a showroom.

There is nothing to reimplement here, and no honest way to guess at it. Ask Steinway.

PETROF — a form into their own records, covering four brands

Petrof offer verification of origin: you give them the brand, the type, the serial number and your email, and they answer from their records. It covers PETROF, WEINBACH, RÖSLER and SCHOLZE. No table is published, so none is reproduced here.

Yamaha — runs a lookup; we could not read their site

Yamaha operate their own piano age lookup. We cannot tell you what it contains, because every Yamaha host we tried refused our request for their robots.txt with HTTP 403 — usa, europe, uk, jp and their download host alike.

When a site will not tell us its crawling rules, we do not crawl it. That is the rule this site works to, and it costs us the single most-searched piano term there is. The link above goes to Yamaha; use theirs.

Bösendorfer — nothing published

Searched across their own domain: no serial table, no lookup tool, no dating page. Bösendorfer publish model and history material and nothing that maps a number to a year. Their contact page is the route.

How the other makers were checked, and what that is worth

Guessing at company domains is how invented facts get written down, so each maker's website was taken from the citation in Wikipedia's own table rather than assumed. 57 of the 222 cite a website; 164 cite none, which is what you would expect of a list where 173 of the companies are defunct and many closed before 1930.

The 53 sites not already covered above were then swept — sitemap first, and where there was no sitemap, a two-level walk from the home page following service, support and FAQ sections. Searching in German, French, Italian, Spanish, Czech and Dutch as well as English, because a German maker does not write “serial”. That is how Grotrian was found: their page is /service/seriennummern/.

And read the swept figure knowing how much was there to read. Of the 41 counted as swept, 27 offered this method fewer than 25 pages to look at, and 23 of those offered nothing from a sitemap at all. A host that answers, allows us, and then has almost nothing to search still counts as swept. It is the number of doors opened, not the number of rooms searched.

What that number is not. “Nothing found” means this method found nothing. A maker who publishes a table inside a PDF, behind a search box, or four clicks from the home page would not be found by it, and would not therefore be publishing nothing. The distinction matters enough to keep saying.

Refused, or unreachable — named rather than counted

Four makers' sites told us not to, and that was honoured: Apollo and Samick carry a blanket Disallow: /; Fritz Dobbert names our crawler specifically; Pearl River returns 403 on robots.txt itself, so their rules could not be read and the answer defaults to no. Nothing was taken from any of them.

Seven more simply did not answer from here — ENSCHU, Lester, Sauter, Schiller, Weber, Wendl & Lung and Willis & Co. That is a failure to reach them, not a finding about them, and it is recorded as unassessed.

A second correction, same cause. Borgato was listed here as not answering. It answers perfectly — the first sweep used one HTTP client and Borgato fails under it while responding to another. Their site is 24 pages and carries no serial table. Two makers have now moved out of “unreachable” for that reason, which is a fair warning about what that row is worth: it measures our reach, not their publishing.

A correction. An earlier version of this page said Mason & Hamlin and Blüthner could not be reached. That was wrong, and it came from trying a domain we had guessed at rather than the one their own entries cite. Both answer perfectly, both were swept — Mason & Hamlin's sitemap, Blüthner's service pages — and neither publishes a serial table that this method could find.

The tables that would answer for the rest are somebody's property

Wikipedia's own article opens by defining the reference work this whole subject runs on: a piano atlas is “a book that contains detailed information about piano manufacturers, often including indexes that can match a piano's serial number to its production date”.

That is the honest position, and the sweep above is what confirms it. The figures for most of these 222 makers are not missing — they exist, and they exist in books that are in copyright and in print. The Pierce Piano Atlas is a commercial publication, not a public reference. Retyping its year-by-year tables onto a web page would be copying a book, and attribution does not fix that.

So this page computes a year only where the maker themselves publishes the figures. Everywhere else it tells you who does hold them, and stops. That is a much shorter page than the ones that will confidently date any piano you type in, and it is the reason to trust it.

The name on the fallboard is not a maker's mark

Wikipedia keeps two separate lists, and the gap between them is the whole problem. List of piano manufacturers holds 222 companies. List of piano brand names holds 1,182, and says of itself that some of those names are for “stencil pianos: instruments manufactured by one company for branding and sale by another”, while others were “licensed or sold by the original company to one or more others”.

You do not need an exotic example. Three makers on this page show it plainly:

A serial from a Weinbach is a Petrof serial. Looking it up in a Weinbach table — if you found one — would be looking it up in the wrong book.

Where to go next

Other instruments on the same footing:

• band instrument serials — brass and woodwind, and what their makers publish
• guitar serial numbers — the largest maker list here

How far this page has got

The pool is 222 manufacturers, taken from Wikipedia's list rather than recalled, split by that article into 49 active and 173 defunct.

The last row is the honest remainder. 164 makers have no website in the source list, and for a firm that closed in 1911 there is unlikely ever to be one — their figures, if they survive at all, are in the atlases. They are listed in piano-makers.txt with the date they were extracted.

The guitar page took 253 makers to reach 91 decoders. Pianos have run the other way, exactly as the first six suggested: the figures mostly exist, and mostly sit in a copyrighted book or a manufacturer's private database. Two makers in 222 publish enough to compute with, and this page would rather say that than pretend otherwise.