We have developed a new method for constructing statistical representations of ensembles of similar shapes that uses particle systems to represent surfaces non parametrically and optimally sample surface point correspondences. We used this method to generate models for two clinical datasets: normal vs. Autistic neurological development. Hypothesis testing on these models using a non parametric permutation test of the Hotelling T-squared metric (including false-discovery-rate (FDR) correction) reveals significant group differences. Colormap indicates the magnitude and direction of the linear discriminant.