In this game, black sac'ed a piece (18...Bxf4) to get an attack going, but white defended well and was arguably better most of the way. However, on move 30, white faced this position [diagram]

click for larger view

The white bishop needn't fear ...exd3 since the black e-pawn is pinned. White might have figured that this e-pawn is attacked 3 times and defended only twice, so he ate it:

30.Bxe4? <poisoned pawn #1> Bf5! <and now it's the bishop's turn to be pinned!>

31.Qc4 <tries to escape with discovered check threat> Nxe4 <regaining his piece>

32.Nf6+? <better is 32.Ne7+, forking K+B with 32...Kh8 33.Nxf5 Qxd2 34.Qxd4 = > Kh8 <not 32...Kf8? 33.Qg8+ (perhaps white was hoping for this blunder?>

33.Ndxe4 gxf6
34.Nd6 Qh5? <better is 34...Qf4 with threats of ...Be4+ and Re1+>

Instead of taking that pawn, white should have played: 30.Nf4 Bf5 31.Qg2 exd3 32.Qxg5 Qxg5 33.Rxg5 g6, remaining up a piece for two pawns (note: weak doubled & isolated pawns), with at least some winning chances.

After 34...Qh5? [diagram]:

click for larger view

White can practically get even with 35.Qf1! (not 35.Nxe8?? Be4+ 36.Rg2 Qd1+ with mate to follow), double-attacking the bishop with the rook now in take. A likely continuation: 35...Re2 (threat: ...Qxh2#) 36.Qf4 Bg6 37.Rg4! with probable draw.

However, white went after another bad (in fact, much worse) pawn:

35.Qxd4?? <poisoned pawn #2> Qf3+!