Abstract: Statistical shape analysis of 3-D scanned human heads provides important information for many applications. Nevertheless, special geometry processing techniques have to be developed for consistently parameterizing scans due to the fact that different scanning projects vary in landmark definition, noise control and other factors. For consistent parameterization, fitting a generic model to each scan has proved to be an effective method. In this paper, improved techniques are presented to solve problems in parameterizing different data sets. Principal Component Analysis (PCA) is thus conducted on the consistently parameterized data sets, and shape variances along principal components are demonstrated. In addition, shape variations analyzed by Independent Component Analysis (ICA) are also presented.