Saddle point optimization

 Is there a way to optimize to the geometry which is a minimum with respect
 to all variables except two for which it should be a maximum
 (ie. saddle point of order of 2). These two variables are some known Z-matrix
 I've looked at saddle, ts, and ef params for opt but they don't seem to
 help with this.
 BTW would this be as simple as choosing "-Grad" direction for all
 but these two for which it would be "+Grad" ?
 Jussi Eloranta
 PS. I'm using Gaussian 92/DFT