This paper introduces an approach to the problem of choosing when to zoom a moving camera so as to follow a designated video surveillance target. Rather than trying to maintain a simple viewing constraint (e.g., target > 10% of image), the potential misrecognition of the target is also used to decide when to zoom. A discrete-event approach is used to develop two models of appearance change as well as a model that represents the viewing constraints for target surveillance. Disagreement between the appearance models is taken to indicate a potential loss of target. Supervisory discrete event control theory is used to automatically construct a controller that selects zoom actions to prevent loss of target. The implementation of this controller is overviewed and results presented.