I noticed when I played the French, that I very often ended up never getting around to castling, especially in the Advance variation, which is the majority of my games.

Playing both sides of the French, it just seems to me that black castling does more to give white a king side attack than to keep black's king safe. Just look at all the textbook examples of the "Greek Gift" sacrifice (Bxh7+), which mostly work because the black pawn on e6 and white pawn on e5 block out black's queen side pieces from aiding in defense.

Obviously, the Exchange Variation and other offbeat responses where black's pawn doesn't stay on e6 are the exception. But in most main lines of the French, it just seems to me like castling is a bad idea.

--Fromper

Heck, sometimes black even castles queenside.

There's no one answer here. Obviously, yes, castling often gives white an attack, even if there is no Bxh7+ sacrifice. And the black's king is often (but not always) fairly safe in the center.

But ...

Black will have a hard time co-ordinating his rooks if he doesn't castle, which can be crucial for pressure down the c-file. The e-file can become very dangerous if black hits at the white center with the thematic ...f6, allowing exf6.

There is no iron-clad rule here.

Compare it to the Sicilian Defense. Should black castle?

In my opinion, sometimes it's okay to castle kingside. In most games I've played against various French defenses, the King is a sitting, camouflaged duck, quacking really loud.
The queenside works more often against me, but it can take up too much time, regardless of a closed position. If you play c5 (why play the French without c5?) there is a draft along the c-file that can prove uncomfortable.
Leaving the king in the middle is often pretty strong but you need to know what you're doing. You need to be alert and decide when (or even if) it will be necessary to make castle, or make a run for it. Early Kf8 plans come in handy as well (the idea coming from the Caro-Kann), a little radical but less commital than actually castling.

Having played the French for almost 25 years, my personal suggestion is to give up altogether this "rule" of delayed castling

Ronaldinho and Perseus are right, sometimes it works, sometimes not. It is true that when white has the cramping e5 pawn and the Bd3/Nf3/g5 batery, there are painful sacs in the air.

But it is also true that quite often Black has the time to play a timely f6 or f5, reducing considerably the danger.

It is correct to say that - with the center closed - often the black king can stay in the center, but this evaluation should always be backed up by other consideration (e.g. mobility of the pieces, especially rook connections)

In general, relying on vague rules such as this one (delayed castling) is VERY dangerous, unless such rules really work in the vast majority of cases, which in my experience is not the case here.

I'd rather forget such a rule and simply try to asses each position as it comes up (as i am sure you have been doing so far )

Perseus, what does this have to do with the Sicilian? From my experience playing white against the Sicilian (not actually playing it as black as I've done with the French), black usually does castle in the Sicilian. The castled king in the Sicilian tends to be a much tougher nut to crack than in the French, especially since there's usually a bishop in front of him at g7.

--Fromper

I mentioned it in respect to castling queenside. In most French defenses you play c5 (boy am I generalizing again!?), thus allowing for some holes through c7 that can prove fatal. Actually, more accurate would be 'when black castles long while his c-pawn has moved'; which happens in French lines with c5, Sicilian lines and the Caro-Kann to name a few. White can't as a rule take sufficient advantage of it because of other considerations. I've found it a simple consideration that when black plays c6 or c5 in combination with 0-0-0 that the castled position can become precarious.
I didn't tie it to kingside castling, which is a different cookie jar as you rightly stated. My apologies for frakking up my post there

(I really should start playing the French again - It's a very interesting opening!)

I'm not sure, but in my book Play the French 3rd Edition by John Watson, I think that in most lines, either white doesn't play e4-e5 (Tarrasch) or black gets rid of white's strong center early on already. Only in lines where white plays f2-f4 (Classical), black can't really "destroy" white's center with f6, but then black is usually ahead in development with moves like Nc6, Qb6 etc. and white hasn't had the time to play Bd3 and Nf3 yet...

But I'm very unsure about the above and I wouldn't take it for granted, because if it were true, there would hardly be any downsides to playing the French at all

I've seen 'logical' undermining operations with f6 in the Steinitz setup that lost the game nearly on the spot.

Good general ideas only fail to specific variations.

The idea of 0-0 or 0-0-0 is King Safety, and mobility of pieces,
linking the rooks.
Looking over French Defense games by GMs it appears that more than several lines Black does not castle. 0-0 or 0-0-0 is dependent on the position.
Reuben Fine:
"The great virtue of the defense is that it combines solidity with hidden resources for the second player. It tempts White to attack, but many of the most tempting attacks turn out to be premature".
The tree of the French defense against 3.Nc3 is
(a) Nf6 (b) Bb4 (Winawer) or (c) d5:e4.
Look over the games by Botvinnik. The French appears in Botvinnik's 100 Games. Also he played it against Smyslov in both matches, and Tal.
Although the Tal match was known for the Nimzovitch variation of the Caro-Kann.
Statistics on The French Defense:
From 1620 to 2007, Chessgames has 2559 games in the database. Since the particular variation is not indicated
in the above threads we have to go with general information.
All in all, White has won 40.8% of the time.
Black 32.3% of the time.
Draws 27.0%.
Black side players:
Alexei Barsov 38 games
Viktor Korchnoi 19 games
Yuri Kruppa 14 games
By Variation: