Classifying and organizing documents in repositories is an active research topic in digital library studies. Manually classifying the large volume of patents and patent applications managed by patent offices is a labor-intensive task. Many previous studies have employed patent contents for patent classification with the aim of automating this process. In this research we propose to use patent citation information, especially the citation network structure information, to address the patent classification problem. We adopt a kernel-based approach and design kernel functions to capture content information and various citation-related information in patents. These kernels. performances are evaluated on a testbed of patents related to nanotechnology. Evaluation results show that our proposed labeled citation graph kernel, which utilized citation network structures, significantly outperforms the kernels that use no citation information or only direct citation information.