One of the soundest and most popular gambits for Black is generally not termed a gambit at all, the Two Knights Defence which arises from 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6.
Most of my recent games as Black in this line have continued with 4.d3, protecting the e4-pawn. Against this I have had some success with a risky plan involving queenside castling and kingside play, while a sounder way of playing actively, as has been recommended a few times on the Chesspublishing.com forum, is to play 4...Bc5 and 5...0-0 and strive for ...d7-d5 (the immediate 4...d5 is dubious because it leaves Black vulnerable to quick attacks down the e-file and the e5-pawn often drops off, but this is less of an issue once Black has castled).
However, this article focuses on some variations following the critical 4.Ng5, where White defends e4 and attacks f7. Although Black can take advantage of White's violation of traditional opening principles (moving a piece twice early in the opening) this is only possible at the cost of material, which leaves the resulting positions dynamically balanced.
I had a recent game in the Ulvestad Variation (4...d5 5.exd5 b5!?) in which Black aims to generate compensation by placing the c8-bishop on the a8-h1 diagonal in many variations, pointing at White's kingside. I don't think it is theoretically as good as 5...Na5, but at the club level it can be very effective as it is easier for White to go wrong (in particular White is unlikely to find the response 6.Bf1! unless he or she has specifically prepared for the line.)
OK, so as is typical at club level, there were quite a few errors. White didn't respond very accurately, and I should have played 9...h6 intending 10.Nf3 e4, before White had the opportunity to get in d2-d3 (which White passed up, and so I took a later opportunity with 12...h6 13.Nf3 e4). The sacrifice 19.Nxh6+ was virtually forced because of the threats to the white queen. The other hiccup was at move 27, when I would have been winning after 27...f6, whereas after my 27...Bf6?, White could have played 28.Re8+! Bxe8 29.Bxf6, which would have forced me to give up an exchange to avoid mate, and would have left me with an extra piece for three pawns, which would have left me only slightly better. I saw this idea during the game after I had played my 27th move but fortunately my opponent didn't spot it.
However the game provides quite a good illustration of the sort of attacking chances that Black gets in these lines. As I say, the only really critical test of 5...b5 is 6.Bf1, which I intend to cover at my chess site in the near future.
However, firstly, I have taken quite a thorough look at the line 4...d5 5.exd5 Nxd5, where Black refuses to part with material. What could be wrong with keeping the material level and then aiming to punish White for moving the f3-knight twice in the opening? Well, the problem for Black is that White doesn't have to back-pedal with the knight on g5, and can happily sacrifice it on f7 in order to bring the black king out into the open.
This line has been analysed extensively by Dan Heisman, and my coverage will be somewhat more lightweight by comparison, but I have located a couple of recent games, one involving the immediate 6.Nxf7 (generally known as the Fegatello or Fried Liver Attack) and the other involving 6.d4, generally known as the Lolli Attack, in which White hopes to sacrifice on f7 under improved circumstances in the next few moves.
6.d4 is the more likely of the two to provide White with a theoretical advantage. There are some tricky lines following 6...Nxd4 7.c3 b5 8.Bxd5 Qxd5, in which Black sacrifices a piece for compensation, though I think White should be able to retain the upper hand. 6...Be6 as played in the illustrative game is probably Black's most secure way of restricting White to just a small advantage, since there are now no knight sacrifices on f7 to worry about, though the game provides a good illustration of how White tends to retain long-term attacking chances.
However, in practice I would still be tempted to wheel out 6.Nxf7 which was always my preference when I played these lines as a junior. The line forces Black's king out into the middle of the board, via 6...Kxf7 7.Qf3+ Ke6 8.Nc3 and now Black's only way to stop White from gaining a near-unstoppable attack is to attack c2 with 8...Ncb4.
In this position, the traditional "book" recommendation of 9.a3, sacrificing the rook on a1, is flawed, as Stefan Bucker and "Master_Om" at the Chesspubishing.com forum have demonstrated. Black grabs the rook on a1, 9...Nxc2+ 10.Kd1 Nxa1, and after the apparently strong 11.Nxd5, plays 11...Qh4. Black is a whole rook up and so can afford to give back material in order to stifle the white attack.
However, after 9.Qe4, 9.0-0 or even 9.Bb3, I am convinced that White has at least sufficient compensation for the sacrificed knight, and the real question is whether or not it is sufficient to give White a theoretical advantage. The illustrative game that I chose continued with one of the most critical lines, 9.Qe4 c6 10.a3 Na6 11.d4 Nac7 12.Bf4 Kf7 13.Bxe5 Be6 14.Qf3+ Kg8.
Black's king has retreated into relative safety but the h8-rook is hemmed in and White has two pawns and a strong centre as compensation for the sacrificed knight. Actually, I feel that the line 12.f4 Kf7 13.fxe5 Be6 14.0-0+ Kg8 may be a slightly improved version of this for White, since White ends up with a half-open f-file and a strong pawn centre with pawns on d4 and e5. But anyway, in the illustrative game White was able to demonstrate sufficient compensation for the piece and Black eventually succumbed to the pressure.
For these reasons I cannot recommend 5...Nxd5 for Black, even though it might theoretically only concede a small disadvantage with best play.