To deal with the problem of Debris' feature extraction with the characteristic of large scale number in description parameter and nonlinear relationship among these parameters, a KPCA-based method is presented. Based on the detailed introduction of KPCA algorithm, a real case was studied for the purpose of constructing a debris automatic recognition sub-system which can be served for diesel engine fault detection and analysis system. The results of experimental research and a comparison with linear PCA demonstrate that KPCA-based approach is feasible and valid for synthesizing nonlinear characteristic parameter and reducing feature dimension.