The objective of this article is to assess the relevance of a statistical method for hyperspectral image classification. We focus on the implementation of a functional method whose main objective is to consider each hyperspectrum as a continuous curve in order to predict its associated class. The implemented functional nonparametric discrimination method is a recently developed technique whose performance are greatly dependent on the choice of a "proximity measure". Behavior in practice of this method has been compared with three more standard others on two sets of hyperspectral data with supervised classification for 50 independent sets using a classification error rate criterion. Experimental results show that this method provides an interesting alternative to conventional methods.