Abstract: A kernel-based improved alternative fuzzy C-means (KIAFCM) clustering algorithm was presented in this paper, which combined the advances of kernel-based learning approach and IAFCM algorithm, and could effectively cluster non-hyper spherical samples, or samples with noise, outliers, etc. The KIAFCM algorithm non-linearly mapped the feature space into the high-dimensional kernel space to improve the clustering performance. Results show that the proposed algorithm is effective.