A novel approach for shape modeling of the corpus callosum (cc) is introduced where the contours of the cc are extracted by image/volume segmentation, and a Bezier curve is used to connect the vertices of the sampled contours, generating a parametric polynomial representation. These polynomials are shown to maintain the characteristics of the original cc, thus are suitable for classification of populations. The Bernstein polynomials are used in fitting the Bezier curves. The coefficients of the Bernstein polynomials are shown to capture the geometric features of the cc, and are able to describe deformations. We use these coefficients, in conjunction with the Fourier Descriptors and other features, to discriminate between autistic and normal brains. The approach is tested on T1-weighted MRI scans of 16 normal and 22 autistic subjects and shows its ability to provide perfect classification, suggesting that the approach is worth investigating on a larger population with the hope of providing early identification and intervention of autism using neuroimaging.