Statistical shape analysis methods are increasingly used in neuroscience and clinical research. A current challenge for methodological research is to perform statistical analysis on large datasets (several hundreds or thousands of subjects). A common approach in morphometry is template-based analysis where one analyzes the deformations that map individuals to a template of the population (Ashburner 1998; Vaillant 2004). The Large Deformation Diffeomorphic Metric Mapping (LDDMM) framework provides a natural setting for quantifying deformations between anatomical shapes. Various methods have been proposed to estimate a template using the LDDMM framework (Durrleman 2008; Glaunès 2006). However, their application to large datasets has remained limited due to their high computational load. We present a fast method for template-based shape analysis in the LDDMM framework. We evaluate the approach on synthetic and real datasets of hippocampal shapes, including a large dataset of 1000 subjects.