Theory recently developed to construct confidence regions based on the parametric bootstrap is applied to add inferential information to graphical displays of sample centroids in canonical variate analysis. Problems of morphometric differentiation among subspecies and species are addressed using numerical resampling procedures.