Abstract: The features involved in modeling of quality prediction and fault diagnosis are in large number, complex and nonlinear, which makes the model compute in higher dimensions and time complexity, as well as a bad accuracy. To solve the above problems, a feature extraction method based on kernel principal component analysis (KPCA) and rough set (RS) was proposed. First, the features were extracted by the KPCA, and then reduced by the RS. Principle and specific implementation steps of the proposed method were introduced. In the end, simulations were carried out based on the operation process of tin bath in a glass manufactory, and SVM based model was established using the data captured in factual production as input. The features extracted from KPCA, RS and the proposed method were input into SVM models. The experimental results were compared with one another, which showed that the features extracted from integrated method of KPCA and RS were better than the others.