Why do you ask?

The formula is:

In: VECTOR V, MATRIX M
Out: VECTOR V

float x,y,z,w;

x = V.x*M._11 + V.y*M._21 + V.z*M._31 + M._41;
y = V.x*M._12 + V.y*M._22 + V.z*M._32 + M._42;
z = V.x*M._13 + V.y*M._23 + V.z*M._33 + M._43;
w = V.x*M._14 + V.y*M._24 + V.z*M._34 + M._44;

V.x = x/w;
V.y = y/w;
V.z = z/w;

If you look in D3DUtil.cpp or D3DMath.cpp you can see Microsoft''s implementation. Of course they may have changed it slightly for D3DX, most likely to support Intel and AMD specific code, but the principle is the same.

WitchLord