A novel classification algorithm of hyperspectral remote sensing image based on nonlinear kernel mapping artificial immune network was proposed. An artificial immune network model was constructed according to natural immune network theory. The training samples of hyperspectral imagery are projected to high feature space with nonlinear kernel function, which improved the sorting method based on similarity in kernel space in artificial immune network. The number of antibodies which are used to recognize training samples is reduced, and the accuracy and efficiency of the algorithm are enhanced. To evaluate the advantage of the proposed algorithm, some other kinds of hyperspectral image classification algorithms were compared with it in the experiment using two hyperspectral image data. Experimental results demonstrated that the proposed algorithm, which acquires higher accuracy and computing speed than traditional hyperspectral image classification algorithms, is a new improved classification algorithm of hyperspectral remote sensing image based on artificial immune network.