Transfer Learning is utilized to the traditional clustering methods to enhance the clustering effect in the target domain with insufficient data. These methods can obtain good performance in handling the linear data. But for the ‘non-linear’ data, they cannot perform effectively. To solve this issue, this paper proposes a novel transfer learning based kernel fuzzy clustering algorithm. In this method, the kernel clustering is first performed to the source data to achieve the important knowledge, i.e. cluster prototypes. Then these cluster prototypes are transferred to guide the data clustering in the target domain. Experiments conducted on the ‘non-linear’ data prove the superiority and efficiency of the proposed method.