Abstract: The principal component analysis is a widely used method for dimensionality reduction of hyperspectral remote sensing images. In task-oriented work, the principal component selection method based on cumulative variance contribution rate is not ideal. To address the problem of principal component selection after principal component analysis transformation, a method of principal component selection based on spatial statistics is proposed. The selection of principal components is performed by calculating the values of the semi-variogram parameter range and partial sill/sill of each principal component. The magnitude of a range is used to judge the range of spatial correlation of each principal component, and the partial sill/sill is used to judge the strength of spatial correlation of each principal component. The simulation proves that the variable range and partial sill/sill can effectively express the range and strength of spatial correlation of hyperspectral remote sensing images. Based on the experiment of real hyperspectral remote sensing images, the empirical threshold of principal component selection is determined from subjective and objective aspects, that is, the range is 2.5, and the partial sill/sill is 0.2. According to the classification results based on the support vector machine algorithm, compared with traditional methods, the principal components with better image quality can be screened by using variable range and partial sill/sill, which can not only achieve the purpose of dimensionality reduction, but also ensure high classification accuracy.