Peter, but sent to the list.

I've not gone through to compare sources, but:

(i) the Economic Census provides a breakdown by capitalization (the two biggest classes, so above \10 oku, account for 25% of employment, and then gives by sector, table 5 Table 5 Enterprises, Establishments and Persons Engaged by Sex, by Capital Size (10 Groups) for Japan, Prefectures and Municipalities, file c05000.
csv).
(ii) ibid by employee size finds Table 2 in three very large files provides info for single and multiple establishment enterprises, giving 49% of regular employees (常用雇用者数) in large (300+) establishments but no info on part-time etc etc so it's less than ideal. This is from initial 2012 data.
(iii) The Chusho Kigyo Hakusho (SME White Paper) gives Econ Census data for the previous 2009 census, with 63% of regular employees in small firms (which seems to be much lower than the only partially released 2012 version). That jumps to 79% if you look at small establishments vs small enterprises.
(iv) the annual Census of Manufacturers provides similar data but not including retail etc. for 2010 only 31% were in 300+ establishments (no enterprise data). there's a similar set of censuses for wholesale and retail, where even fewer are in large establishments.

Since you've looked at them, I've not looked at the labor-related ones.

My sense from past work is that (a) the overall picture doesn't vary much while (b) definitions vary a lot and
(c) that's even more true if you move to international comparisons. I used to say that 2/3rds of Japanese worked for "small" firms (300 employees as cutoff) while 2/3rds of people in the US worked for large (500+?). That's based on old data, Japan has seen a small decline, but I don't know about the US.

Unless what you're doing demands otherwise, I would urge you to stick to such "round" numbers rather than giving 2 significant digits, which implies precision beyond what the data allow, particularly in comparative work. Because the Economic Census (at least for the data so far available for 2012) doesn't include paato and hakenshain and so on, you can check the labor survey data against the Economic Census data (the labor surveys give "regular" employment -- are levels
similar?) but if there's no huge divergence, stick to it. It also provides better time series coverage.

mike smitka