Hyperspectral imagery classification, taking into account spectral and spatial features, is a promising task in remote sensing field. In this paper, the incorporation of second order statistics in hyperspectral data classification using support vector machines is proposed. The effect of using the semivariance geostatistic as a spatial feature rather than first order statistics (mean and standard deviation) is tested. The overall classification accuracy is evaluated for the AVIRIS Indian Pines-1992 benchmark data set. Empirical results show that the proposed approach gives better performance than the method based on first order statistics