Quote:

You start with the unit cube, and transform it by the inverse clip matrix and then extract the planes from the new points. Enjoy.

Right, we're on the same wavelength. I think that what I mentioned above is similar, except it starts one step ahead: instead of starting with the unit cube, it starts out with that cube transformed such that it represents your field of view (the projection matrix in OpenGL I believe), then multiplying it by the modelview/worldview matrix brings it into worldspace.