Abstract: The Support Vector Machines (SVM) are learning supervised techniques developed by Vapnik. Their learning, has its roots in the statistical theories with discrimination based on a linear separation in an adequate dimension space. The change of dimension is done through kernel function, which must be chosen from several. In order to evaluate the contribution of the choice of kernel on the SVMs performance, we conducted a classification of a satellite image representing the western region of ORAN, in ALGERIA, with varying kernels and their parameters.