Abstract: The mass composition of high energy cosmic rays above 10(17) eV is a crucial issue to solve some open questions in astrophysics such as the acceleration and propagation mechanisms. Unfortunately, the standard procedures to identify the primary particle of a cosmic ray shower have low efficiency mainly due to large fluctuations in the shower development. We present a statistical method for composition studies based on several measurable features of the longitudinal development of the CR shower such as N-max, X-max, asymmetry, skewness and kurtosis. Principal component analysis was used to evaluate the relevance of each parameter in the representation of the overall shower features and a linear discriminant analysis (LDA) was used to combine the different parameters to maximize the discrimination between different particle showers. LDA provides a separation between primary gammas, proton and iron nuclei better than the procedures based on X-max, only. The method proposed herein was successfully tested in the energy range from 10(17) to 10(20) eV even when I imitations of shower track length were included in order to simulate the field of view of fluorescence telescopes Ver menos