This paper will describe a simple technique that can be used to generalize the target classification algorithms employed by passive midcourse sensors for strategic defense. Most discrimination algorithm evaluations have assumed a fixed engagement geometry (target location/orientation, sensor location, sun and earth angles). Pattern classifiers are trained and tested in that geometry and therefore are not fully applicable in a full scale engagement. By training on the full range of potential engagements, the important class signature dependencies can be stored in an expanded mean vector and covariance matrix . Then through standard statistical techniques, the mean and covariance can be properly conditioned to the geometry applicable to a particular track file. This paper demonstrates that this approach is capable of adapting discrimination algorithms to a general scenario without significant loss in classification accuracy.